Medix-AI/coqui-tts
Medix-AI
/
coqui-tts
mirror of https://github.com/coqui-ai/TTS.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'dev' into fix-improvements/adjust-speech-rate-or-speed

This commit is contained in:
Işık 2024-12-30 00:36:41 +03:00 committed by GitHub
parent 26128be422 98080e282c
commit ed1563b132
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
340 changed files with 5196 additions and 7573 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
.cardboardlint.yml
View File

@ -1,5 +0,0 @@
linters:
- pylint:
# pylintrc: pylintrc
filefilter: ['- test_*.py', '+ *.py', '- *.npy']
# exclude:

2
.github/ISSUE_TEMPLATE/bug_report.yaml vendored
View File

@ -59,7 +59,7 @@ body:
You can either run `TTS/bin/collect_env_info.py`
```bash
wget https://raw.githubusercontent.com/coqui-ai/TTS/main/TTS/bin/collect_env_info.py
wget https://raw.githubusercontent.com/idiap/coqui-ai-TTS/main/TTS/bin/collect_env_info.py
python collect_env_info.py
```

4
.github/ISSUE_TEMPLATE/config.yml vendored
View File

@ -1,8 +1,8 @@
blank_issues_enabled: false
contact_links:
- name: CoquiTTS GitHub Discussions
url: https://github.com/coqui-ai/TTS/discussions
url: https://github.com/idiap/coqui-ai-TTS/discussions
about: Please ask and answer questions here.
- name: Coqui Security issue disclosure
url: mailto:info@coqui.ai
url: mailto:enno.hermann@gmail.com
about: Please report security vulnerabilities here.

8
.github/PR_TEMPLATE.md vendored
View File

@ -5,11 +5,3 @@ Welcome to the 🐸TTS project! We are excited to see your interest, and appreci
This repository is governed by the Contributor Covenant Code of Conduct. For more details, see the [CODE_OF_CONDUCT.md](CODE_OF_CONDUCT.md) file.
In order to make a good pull request, please see our [CONTRIBUTING.md](CONTRIBUTING.md) file.
Before accepting your pull request, you will be asked to sign a [Contributor License Agreement](https://cla-assistant.io/coqui-ai/TTS).
This [Contributor License Agreement](https://cla-assistant.io/coqui-ai/TTS):
- Protects you, Coqui, and the users of the code.
- Does not change your rights to use your contributions for any purpose.
- Does not change the license of the 🐸TTS project. It just makes the terms of your contribution clearer and lets us know you are OK to contribute.

11
.github/actions/setup-uv/action.yml vendored Normal file
View File

@ -0,0 +1,11 @@
name: Setup uv
runs:
using: 'composite'
steps:
- name: Install uv
uses: astral-sh/setup-uv@v4
with:
version: "0.5.4"
enable-cache: true
cache-dependency-glob: "**/pyproject.toml"

1
.github/stale.yml vendored
View File

@ -15,4 +15,3 @@ markComment: >
for your contributions. You might also look our discussion channels.
# Comment to post when closing a stale issue. Set to `false` to disable
closeComment: false

51
.github/workflows/aux_tests.yml vendored
View File

@ -1,51 +0,0 @@
name: aux-tests
on:
push:
branches:
- main
pull_request:
types: [opened, synchronize, reopened]
jobs:
check_skip:
runs-on: ubuntu-latest
if: "! contains(github.event.head_commit.message, '[ci skip]')"
steps:
- run: echo "${{ github.event.head_commit.message }}"
test:
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
python-version: [3.9, "3.10", "3.11"]
experimental: [false]
steps:
- uses: actions/checkout@v3
- name: Set up Python ${{ matrix.python-version }}
uses: actions/setup-python@v4
with:
python-version: ${{ matrix.python-version }}
architecture: x64
cache: 'pip'
cache-dependency-path: 'requirements*'
- name: check OS
run: cat /etc/os-release
- name: set ENV
run: export TRAINER_TELEMETRY=0
- name: Install dependencies
run: |
sudo apt-get update
sudo apt-get install -y git make gcc
make system-deps
- name: Install/upgrade Python setup deps
run: python3 -m pip install --upgrade pip setuptools wheel
- name: Replace scarf urls
run: |
sed -i 's/https:\/\/coqui.gateway.scarf.sh\//https:\/\/github.com\/coqui-ai\/TTS\/releases\/download\//g' TTS/.models.json
- name: Install TTS
run: |
python3 -m pip install .[all]
python3 setup.py egg_info
- name: Unit tests
run: make test_aux

51
.github/workflows/data_tests.yml vendored
View File

@ -1,51 +0,0 @@
name: data-tests
on:
push:
branches:
- main
pull_request:
types: [opened, synchronize, reopened]
jobs:
check_skip:
runs-on: ubuntu-latest
if: "! contains(github.event.head_commit.message, '[ci skip]')"
steps:
- run: echo "${{ github.event.head_commit.message }}"
test:
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
python-version: [3.9, "3.10", "3.11"]
experimental: [false]
steps:
- uses: actions/checkout@v3
- name: Set up Python ${{ matrix.python-version }}
uses: actions/setup-python@v4
with:
python-version: ${{ matrix.python-version }}
architecture: x64
cache: 'pip'
cache-dependency-path: 'requirements*'
- name: check OS
run: cat /etc/os-release
- name: set ENV
run: export TRAINER_TELEMETRY=0
- name: Install dependencies
run: |
sudo apt-get update
sudo apt-get install -y --no-install-recommends git make gcc
make system-deps
- name: Install/upgrade Python setup deps
run: python3 -m pip install --upgrade pip setuptools wheel
- name: Replace scarf urls
run: |
sed -i 's/https:\/\/coqui.gateway.scarf.sh\//https:\/\/github.com\/coqui-ai\/TTS\/releases\/download\//g' TTS/.models.json
- name: Install TTS
run: |
python3 -m pip install .[all]
python3 setup.py egg_info
- name: Unit tests
run: make data_tests

65
.github/workflows/docker.yaml vendored
View File

@ -10,7 +10,7 @@ on:
jobs:
docker-build:
name: "Build and push Docker image"
runs-on: ubuntu-20.04
runs-on: ubuntu-latest
strategy:
matrix:
arch: ["amd64"]
@ -18,7 +18,7 @@ jobs:
- "nvidia/cuda:11.8.0-base-ubuntu22.04" # GPU enabled
- "python:3.10.8-slim" # CPU only
steps:
- uses: actions/checkout@v2
- uses: actions/checkout@v4
- name: Log in to the Container registry
uses: docker/login-action@v1
with:
@ -29,11 +29,11 @@ jobs:
id: compute-tag
run: |
set -ex
base="ghcr.io/coqui-ai/tts"
base="ghcr.io/idiap/coqui-tts"
tags="" # PR build
if [[ ${{ matrix.base }} = "python:3.10.8-slim" ]]; then
base="ghcr.io/coqui-ai/tts-cpu"
base="ghcr.io/idiap/coqui-tts-cpu"
fi
if [[ "${{ startsWith(github.ref, 'refs/heads/') }}" = "true" ]]; then
@ -42,7 +42,7 @@ jobs:
branch=${github_ref#*refs/heads/} # strip prefix to get branch name
tags="${base}:${branch},${base}:${{ github.sha }},"
elif [[ "${{ startsWith(github.ref, 'refs/tags/') }}" = "true" ]]; then
VERSION="v$(cat TTS/VERSION)"
VERSION="v$(grep -m 1 version pyproject.toml | grep -P '\d+\.\d+\.\d+' -o)"
if [[ "${{ github.ref }}" != "refs/tags/${VERSION}" ]]; then
echo "Pushed tag does not match VERSION file. Aborting push."
exit 1
@ -63,3 +63,58 @@ jobs:
push: ${{ github.event_name == 'push' }}
build-args: "BASE=${{ matrix.base }}"
tags: ${{ steps.compute-tag.outputs.tags }}
docker-dev-build:
name: "Build the development Docker image"
runs-on: ubuntu-latest
strategy:
matrix:
arch: ["amd64"]
base:
- "nvidia/cuda:11.8.0-base-ubuntu22.04" # GPU enabled
steps:
- uses: actions/checkout@v4
- name: Log in to the Container registry
uses: docker/login-action@v1
with:
registry: ghcr.io
username: ${{ github.actor }}
password: ${{ secrets.GITHUB_TOKEN }}
- name: Compute Docker tags, check VERSION file matches tag
id: compute-tag
run: |
set -ex
base="ghcr.io/idiap/coqui-tts-dev"
tags="" # PR build
if [[ ${{ matrix.base }} = "python:3.10.8-slim" ]]; then
base="ghcr.io/idiap/coqui-tts-dev-cpu"
fi
if [[ "${{ startsWith(github.ref, 'refs/heads/') }}" = "true" ]]; then
# Push to branch
github_ref="${{ github.ref }}"
branch=${github_ref#*refs/heads/} # strip prefix to get branch name
tags="${base}:${branch},${base}:${{ github.sha }},"
elif [[ "${{ startsWith(github.ref, 'refs/tags/') }}" = "true" ]]; then
VERSION="v$(grep -m 1 version pyproject.toml | grep -P '\d+\.\d+\.\d+' -o)"
if [[ "${{ github.ref }}" != "refs/tags/${VERSION}" ]]; then
echo "Pushed tag does not match VERSION file. Aborting push."
exit 1
fi
tags="${base}:${VERSION},${base}:latest,${base}:${{ github.sha }}"
fi
echo "::set-output name=tags::${tags}"
- name: Set up QEMU
uses: docker/setup-qemu-action@v1
- name: Set up Docker Buildx
id: buildx
uses: docker/setup-buildx-action@v1
- name: Build and push
uses: docker/build-push-action@v2
with:
context: .
file: dockerfiles/Dockerfile.dev
platforms: linux/${{ matrix.arch }}
push: false
build-args: "BASE=${{ matrix.base }}"
tags: ${{ steps.compute-tag.outputs.tags }}

53
.github/workflows/inference_tests.yml vendored
View File

@ -1,53 +0,0 @@
name: inference_tests
on:
push:
branches:
- main
pull_request:
types: [opened, synchronize, reopened]
jobs:
check_skip:
runs-on: ubuntu-latest
if: "! contains(github.event.head_commit.message, '[ci skip]')"
steps:
- run: echo "${{ github.event.head_commit.message }}"
test:
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
python-version: [3.9, "3.10", "3.11"]
experimental: [false]
steps:
- uses: actions/checkout@v3
- name: Set up Python ${{ matrix.python-version }}
uses: actions/setup-python@v4
with:
python-version: ${{ matrix.python-version }}
architecture: x64
cache: 'pip'
cache-dependency-path: 'requirements*'
- name: check OS
run: cat /etc/os-release
- name: set ENV
run: |
export TRAINER_TELEMETRY=0
- name: Install dependencies
run: |
sudo apt-get update
sudo apt-get install -y --no-install-recommends git make gcc
sudo apt-get install espeak-ng
make system-deps
- name: Install/upgrade Python setup deps
run: python3 -m pip install --upgrade pip setuptools wheel
- name: Replace scarf urls
run: |
sed -i 's/https:\/\/coqui.gateway.scarf.sh\//https:\/\/github.com\/coqui-ai\/TTS\/releases\/download\//g' TTS/.models.json
- name: Install TTS
run: |
python3 -m pip install .[all]
python3 setup.py egg_info
- name: Unit tests
run: make inference_tests

98
.github/workflows/pypi-release.yml vendored
View File

@ -7,88 +7,48 @@ defaults:
shell:
bash
jobs:
build-sdist:
runs-on: ubuntu-20.04
build:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v3
- uses: actions/checkout@v4
- name: Setup uv
uses: ./.github/actions/setup-uv
- name: Verify tag matches version
run: |
set -ex
version=$(cat TTS/VERSION)
version=$(grep -m 1 version pyproject.toml | grep -P '\d+\.\d+\.\d+' -o)
tag="${GITHUB_REF/refs\/tags\/}"
if [[ "v$version" != "$tag" ]]; then
exit 1
fi
- uses: actions/setup-python@v2
with:
python-version: 3.9
- run: |
python -m pip install -U pip setuptools wheel build
- run: |
python -m build
- run: |
pip install dist/*.tar.gz
- uses: actions/upload-artifact@v2
with:
name: sdist
path: dist/*.tar.gz
build-wheels:
runs-on: ubuntu-20.04
strategy:
matrix:
python-version: ["3.9", "3.10", "3.11"]
steps:
- uses: actions/checkout@v3
- uses: actions/setup-python@v2
with:
python-version: ${{ matrix.python-version }}
- name: Install pip requirements
- name: Set up Python
run: uv python install 3.12
- name: Build sdist and wheel
run: uv build
- name: Test installation of sdist and wheel
run: |
python -m pip install -U pip setuptools wheel build
python -m pip install -r requirements.txt
- name: Setup and install manylinux1_x86_64 wheel
run: |
python setup.py bdist_wheel --plat-name=manylinux1_x86_64
python -m pip install dist/*-manylinux*.whl
- uses: actions/upload-artifact@v2
uv venv --no-project
uv pip install dist/*.tar.gz
uv pip install dist/*.whl
- uses: actions/upload-artifact@v4
with:
name: wheel-${{ matrix.python-version }}
path: dist/*-manylinux*.whl
name: build
path: dist/*
publish-artifacts:
runs-on: ubuntu-20.04
needs: [build-sdist, build-wheels]
name: Publish to PyPI
runs-on: ubuntu-latest
needs: [build]
environment:
name: release
url: https://pypi.org/p/coqui-tts
permissions:
id-token: write
steps:
- run: |
mkdir dist
- uses: actions/download-artifact@v2
- uses: actions/download-artifact@v4
with:
name: "sdist"
path: "dist/"
- uses: actions/download-artifact@v2
with:
name: "wheel-3.9"
path: "dist/"
- uses: actions/download-artifact@v2
with:
name: "wheel-3.10"
path: "dist/"
- uses: actions/download-artifact@v2
with:
name: "wheel-3.11"
path: "dist/"
name: build
- run: |
ls -lh dist/
- name: Setup PyPI config
run: |
cat << EOF > ~/.pypirc
[pypi]
username=__token__
password=${{ secrets.PYPI_TOKEN }}
EOF
- uses: actions/setup-python@v2
with:
python-version: 3.9
- run: |
python -m pip install twine
- run: |
twine upload --repository pypi dist/*
- name: Publish package distributions to PyPI
uses: pypa/gh-action-pypi-publish@release/v1

37
.github/workflows/style_check.yml vendored
View File

@ -7,40 +7,17 @@ on:
pull_request:
types: [opened, synchronize, reopened]
jobs:
check_skip:
runs-on: ubuntu-latest
if: "! contains(github.event.head_commit.message, '[ci skip]')"
steps:
- run: echo "${{ github.event.head_commit.message }}"
test:
lint:
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
python-version: [3.9]
experimental: [false]
steps:
- uses: actions/checkout@v3
- uses: actions/checkout@v4
- name: Setup uv
uses: ./.github/actions/setup-uv
- name: Set up Python ${{ matrix.python-version }}
uses: actions/setup-python@v4
with:
python-version: ${{ matrix.python-version }}
architecture: x64
cache: 'pip'
cache-dependency-path: 'requirements*'
- name: check OS
run: cat /etc/os-release
- name: Install dependencies
run: |
sudo apt-get update
sudo apt-get install -y git make gcc
make system-deps
- name: Install/upgrade Python setup deps
run: python3 -m pip install --upgrade pip setuptools wheel
- name: Install TTS
run: |
python3 -m pip install .[all]
python3 setup.py egg_info
- name: Style check
run: make style
run: uv python install ${{ matrix.python-version }}
- name: Lint check
run: make lint

127
.github/workflows/tests.yml vendored Normal file
View File

@ -0,0 +1,127 @@
name: test
on:
push:
branches:
- main
pull_request:
types: [opened, synchronize, reopened]
workflow_dispatch:
inputs:
trainer_branch:
description: "Branch of Trainer to test"
required: false
default: "main"
coqpit_branch:
description: "Branch of Coqpit to test"
required: false
default: "main"
jobs:
unit:
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
python-version: [3.9, "3.10", "3.11", "3.12"]
subset: ["data_tests", "inference_tests", "test_aux", "test_text"]
steps:
- uses: actions/checkout@v4
- name: Setup uv
uses: ./.github/actions/setup-uv
- name: Set up Python ${{ matrix.python-version }}
run: uv python install ${{ matrix.python-version }}
- name: Install Espeak
if: contains(fromJSON('["inference_tests", "test_text"]'), matrix.subset)
run: |
sudo apt-get update
sudo apt-get install espeak espeak-ng
- name: Install dependencies
run: |
sudo apt-get update
sudo apt-get install -y --no-install-recommends git make gcc
make system-deps
- name: Install custom Trainer and/or Coqpit if requested
run: |
if [[ -n "${{ github.event.inputs.trainer_branch }}" ]]; then
uv add git+https://github.com/idiap/coqui-ai-Trainer --branch ${{ github.event.inputs.trainer_branch }}
fi
if [[ -n "${{ github.event.inputs.coqpit_branch }}" ]]; then
uv add git+https://github.com/idiap/coqui-ai-coqpit --branch ${{ github.event.inputs.coqpit_branch }}
fi
- name: Unit tests
run: |
resolution=highest
if [ "${{ matrix.python-version }}" == "3.9" ]; then
resolution=lowest-direct
fi
uv run --resolution=$resolution --extra server --extra languages make ${{ matrix.subset }}
- name: Upload coverage data
uses: actions/upload-artifact@v4
with:
include-hidden-files: true
name: coverage-data-${{ matrix.subset }}-${{ matrix.python-version }}
path: .coverage.*
if-no-files-found: ignore
integration:
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
python-version: ["3.9", "3.12"]
subset: ["test_tts", "test_tts2", "test_vocoder", "test_xtts", "test_zoo0", "test_zoo1", "test_zoo2"]
steps:
- uses: actions/checkout@v4
- name: Setup uv
uses: ./.github/actions/setup-uv
- name: Set up Python ${{ matrix.python-version }}
run: uv python install ${{ matrix.python-version }}
- name: Install Espeak
if: contains(fromJSON('["test_tts", "test_tts2", "test_xtts", "test_zoo0", "test_zoo1", "test_zoo2"]'), matrix.subset)
run: |
sudo apt-get update
sudo apt-get install espeak espeak-ng
- name: Install dependencies
run: |
sudo apt-get update
sudo apt-get install -y --no-install-recommends git make gcc
make system-deps
- name: Install custom Trainer and/or Coqpit if requested
run: |
if [[ -n "${{ github.event.inputs.trainer_branch }}" ]]; then
uv add git+https://github.com/idiap/coqui-ai-Trainer --branch ${{ github.event.inputs.trainer_branch }}
fi
if [[ -n "${{ github.event.inputs.coqpit_branch }}" ]]; then
uv add git+https://github.com/idiap/coqui-ai-coqpit --branch ${{ github.event.inputs.coqpit_branch }}
fi
- name: Integration tests
run: |
resolution=highest
if [ "${{ matrix.python-version }}" == "3.9" ]; then
resolution=lowest-direct
fi
uv run --resolution=$resolution --extra server --extra languages make ${{ matrix.subset }}
- name: Upload coverage data
uses: actions/upload-artifact@v4
with:
include-hidden-files: true
name: coverage-data-${{ matrix.subset }}-${{ matrix.python-version }}
path: .coverage.*
if-no-files-found: ignore
coverage:
if: always()
needs: [unit, integration]
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- name: Setup uv
uses: ./.github/actions/setup-uv
- uses: actions/download-artifact@v4
with:
pattern: coverage-data-*
merge-multiple: true
- name: Combine coverage
run: |
uv python install
uvx coverage combine
uvx coverage html --skip-covered --skip-empty
uvx coverage report --format=markdown >> $GITHUB_STEP_SUMMARY

50
.github/workflows/text_tests.yml vendored
View File

@ -1,50 +0,0 @@
name: text-tests
on:
push:
branches:
- main
pull_request:
types: [opened, synchronize, reopened]
jobs:
check_skip:
runs-on: ubuntu-latest
if: "! contains(github.event.head_commit.message, '[ci skip]')"
steps:
- run: echo "${{ github.event.head_commit.message }}"
test:
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
python-version: [3.9, "3.10", "3.11"]
experimental: [false]
steps:
- uses: actions/checkout@v3
- name: Set up Python ${{ matrix.python-version }}
uses: actions/setup-python@v4
with:
python-version: ${{ matrix.python-version }}
architecture: x64
cache: 'pip'
cache-dependency-path: 'requirements*'
- name: check OS
run: cat /etc/os-release
- name: set ENV
run: export TRAINER_TELEMETRY=0
- name: Install dependencies
run: |
sudo apt-get update
sudo apt-get install -y --no-install-recommends git make gcc
sudo apt-get install espeak
sudo apt-get install espeak-ng
make system-deps
- name: Install/upgrade Python setup deps
run: python3 -m pip install --upgrade pip setuptools wheel
- name: Install TTS
run: |
python3 -m pip install .[all]
python3 setup.py egg_info
- name: Unit tests
run: make test_text

53
.github/workflows/tts_tests.yml vendored
View File

@ -1,53 +0,0 @@
name: tts-tests
on:
push:
branches:
- main
pull_request:
types: [opened, synchronize, reopened]
jobs:
check_skip:
runs-on: ubuntu-latest
if: "! contains(github.event.head_commit.message, '[ci skip]')"
steps:
- run: echo "${{ github.event.head_commit.message }}"
test:
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
python-version: [3.9, "3.10", "3.11"]
experimental: [false]
steps:
- uses: actions/checkout@v3
- name: Set up Python ${{ matrix.python-version }}
uses: actions/setup-python@v4
with:
python-version: ${{ matrix.python-version }}
architecture: x64
cache: 'pip'
cache-dependency-path: 'requirements*'
- name: check OS
run: cat /etc/os-release
- name: set ENV
run: export TRAINER_TELEMETRY=0
- name: Install dependencies
run: |
sudo apt-get update
sudo apt-get install -y --no-install-recommends git make gcc
sudo apt-get install espeak
sudo apt-get install espeak-ng
make system-deps
- name: Install/upgrade Python setup deps
run: python3 -m pip install --upgrade pip setuptools wheel
- name: Replace scarf urls
run: |
sed -i 's/https:\/\/coqui.gateway.scarf.sh\//https:\/\/github.com\/coqui-ai\/TTS\/releases\/download\//g' TTS/.models.json
- name: Install TTS
run: |
python3 -m pip install .[all]
python3 setup.py egg_info
- name: Unit tests
run: make test_tts

53
.github/workflows/tts_tests2.yml vendored
View File

@ -1,53 +0,0 @@
name: tts-tests2
on:
push:
branches:
- main
pull_request:
types: [opened, synchronize, reopened]
jobs:
check_skip:
runs-on: ubuntu-latest
if: "! contains(github.event.head_commit.message, '[ci skip]')"
steps:
- run: echo "${{ github.event.head_commit.message }}"
test:
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
python-version: [3.9, "3.10", "3.11"]
experimental: [false]
steps:
- uses: actions/checkout@v3
- name: Set up Python ${{ matrix.python-version }}
uses: actions/setup-python@v4
with:
python-version: ${{ matrix.python-version }}
architecture: x64
cache: 'pip'
cache-dependency-path: 'requirements*'
- name: check OS
run: cat /etc/os-release
- name: set ENV
run: export TRAINER_TELEMETRY=0
- name: Install dependencies
run: |
sudo apt-get update
sudo apt-get install -y --no-install-recommends git make gcc
sudo apt-get install espeak
sudo apt-get install espeak-ng
make system-deps
- name: Install/upgrade Python setup deps
run: python3 -m pip install --upgrade pip setuptools wheel
- name: Replace scarf urls
run: |
sed -i 's/https:\/\/coqui.gateway.scarf.sh\//https:\/\/github.com\/coqui-ai\/TTS\/releases\/download\//g' TTS/.models.json
- name: Install TTS
run: |
python3 -m pip install .[all]
python3 setup.py egg_info
- name: Unit tests
run: make test_tts2

48
.github/workflows/vocoder_tests.yml vendored
View File

@ -1,48 +0,0 @@
name: vocoder-tests
on:
push:
branches:
- main
pull_request:
types: [opened, synchronize, reopened]
jobs:
check_skip:
runs-on: ubuntu-latest
if: "! contains(github.event.head_commit.message, '[ci skip]')"
steps:
- run: echo "${{ github.event.head_commit.message }}"
test:
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
python-version: [3.9, "3.10", "3.11"]
experimental: [false]
steps:
- uses: actions/checkout@v3
- name: Set up Python ${{ matrix.python-version }}
uses: actions/setup-python@v4
with:
python-version: ${{ matrix.python-version }}
architecture: x64
cache: 'pip'
cache-dependency-path: 'requirements*'
- name: check OS
run: cat /etc/os-release
- name: set ENV
run: export TRAINER_TELEMETRY=0
- name: Install dependencies
run: |
sudo apt-get update
sudo apt-get install -y git make gcc
make system-deps
- name: Install/upgrade Python setup deps
run: python3 -m pip install --upgrade pip setuptools wheel
- name: Install TTS
run: |
python3 -m pip install .[all]
python3 setup.py egg_info
- name: Unit tests
run: make test_vocoder

53
.github/workflows/xtts_tests.yml vendored
View File

@ -1,53 +0,0 @@
name: xtts-tests
on:
push:
branches:
- main
pull_request:
types: [opened, synchronize, reopened]
jobs:
check_skip:
runs-on: ubuntu-latest
if: "! contains(github.event.head_commit.message, '[ci skip]')"
steps:
- run: echo "${{ github.event.head_commit.message }}"
test:
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
python-version: [3.9, "3.10", "3.11"]
experimental: [false]
steps:
- uses: actions/checkout@v3
- name: Set up Python ${{ matrix.python-version }}
uses: actions/setup-python@v4
with:
python-version: ${{ matrix.python-version }}
architecture: x64
cache: 'pip'
cache-dependency-path: 'requirements*'
- name: check OS
run: cat /etc/os-release
- name: set ENV
run: export TRAINER_TELEMETRY=0
- name: Install dependencies
run: |
sudo apt-get update
sudo apt-get install -y --no-install-recommends git make gcc
sudo apt-get install espeak
sudo apt-get install espeak-ng
make system-deps
- name: Install/upgrade Python setup deps
run: python3 -m pip install --upgrade pip setuptools wheel
- name: Replace scarf urls
run: |
sed -i 's/https:\/\/coqui.gateway.scarf.sh\//https:\/\/github.com\/coqui-ai\/TTS\/releases\/download\//g' TTS/.models.json
- name: Install TTS
run: |
python3 -m pip install .[all]
python3 setup.py egg_info
- name: Unit tests
run: make test_xtts

54
.github/workflows/zoo_tests0.yml vendored
View File

@ -1,54 +0,0 @@
name: zoo-tests-0
on:
push:
branches:
- main
pull_request:
types: [opened, synchronize, reopened]
jobs:
check_skip:
runs-on: ubuntu-latest
if: "! contains(github.event.head_commit.message, '[ci skip]')"
steps:
- run: echo "${{ github.event.head_commit.message }}"
test:
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
python-version: [3.9, "3.10", "3.11"]
experimental: [false]
steps:
- uses: actions/checkout@v3
- name: Set up Python ${{ matrix.python-version }}
uses: actions/setup-python@v4
with:
python-version: ${{ matrix.python-version }}
architecture: x64
cache: 'pip'
cache-dependency-path: 'requirements*'
- name: check OS
run: cat /etc/os-release
- name: set ENV
run: export TRAINER_TELEMETRY=0
- name: Install dependencies
run: |
sudo apt-get update
sudo apt-get install -y git make gcc
sudo apt-get install espeak espeak-ng
make system-deps
- name: Install/upgrade Python setup deps
run: python3 -m pip install --upgrade pip setuptools wheel
- name: Replace scarf urls
run: |
sed -i 's/https:\/\/coqui.gateway.scarf.sh\//https:\/\/github.com\/coqui-ai\/TTS\/releases\/download\//g' TTS/.models.json
- name: Install TTS
run: |
python3 -m pip install .[all]
python3 setup.py egg_info
- name: Unit tests
run: |
nose2 -F -v -B TTS tests.zoo_tests.test_models.test_models_offset_0_step_3
nose2 -F -v -B TTS tests.zoo_tests.test_models.test_voice_conversion

53
.github/workflows/zoo_tests1.yml vendored
View File

@ -1,53 +0,0 @@
name: zoo-tests-1
on:
push:
branches:
- main
pull_request:
types: [opened, synchronize, reopened]
jobs:
check_skip:
runs-on: ubuntu-latest
if: "! contains(github.event.head_commit.message, '[ci skip]')"
steps:
- run: echo "${{ github.event.head_commit.message }}"
test:
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
python-version: [3.9, "3.10", "3.11"]
experimental: [false]
steps:
- uses: actions/checkout@v3
- name: Set up Python ${{ matrix.python-version }}
uses: actions/setup-python@v4
with:
python-version: ${{ matrix.python-version }}
architecture: x64
cache: 'pip'
cache-dependency-path: 'requirements*'
- name: check OS
run: cat /etc/os-release
- name: set ENV
run: export TRAINER_TELEMETRY=0
- name: Install dependencies
run: |
sudo apt-get update
sudo apt-get install -y git make gcc
sudo apt-get install espeak espeak-ng
make system-deps
- name: Install/upgrade Python setup deps
run: python3 -m pip install --upgrade pip setuptools wheel
- name: Replace scarf urls
run: |
sed -i 's/https:\/\/coqui.gateway.scarf.sh\/hf\/bark\//https:\/\/huggingface.co\/erogol\/bark\/resolve\/main\//g' TTS/.models.json
sed -i 's/https:\/\/coqui.gateway.scarf.sh\//https:\/\/github.com\/coqui-ai\/TTS\/releases\/download\//g' TTS/.models.json
- name: Install TTS
run: |
python3 -m pip install .[all]
python3 setup.py egg_info
- name: Unit tests
run: nose2 -F -v -B --with-coverage --coverage TTS tests.zoo_tests.test_models.test_models_offset_1_step_3

52
.github/workflows/zoo_tests2.yml vendored
View File

@ -1,52 +0,0 @@
name: zoo-tests-2
on:
push:
branches:
- main
pull_request:
types: [opened, synchronize, reopened]
jobs:
check_skip:
runs-on: ubuntu-latest
if: "! contains(github.event.head_commit.message, '[ci skip]')"
steps:
- run: echo "${{ github.event.head_commit.message }}"
test:
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
python-version: [3.9, "3.10", "3.11"]
experimental: [false]
steps:
- uses: actions/checkout@v3
- name: Set up Python ${{ matrix.python-version }}
uses: actions/setup-python@v4
with:
python-version: ${{ matrix.python-version }}
architecture: x64
cache: 'pip'
cache-dependency-path: 'requirements*'
- name: check OS
run: cat /etc/os-release
- name: set ENV
run: export TRAINER_TELEMETRY=0
- name: Install dependencies
run: |
sudo apt-get update
sudo apt-get install -y git make gcc
sudo apt-get install espeak espeak-ng
make system-deps
- name: Install/upgrade Python setup deps
run: python3 -m pip install --upgrade pip setuptools wheel
- name: Replace scarf urls
run: |
sed -i 's/https:\/\/coqui.gateway.scarf.sh\//https:\/\/github.com\/coqui-ai\/TTS\/releases\/download\//g' TTS/.models.json
- name: Install TTS
run: |
python3 -m pip install .[all]
python3 setup.py egg_info
- name: Unit tests
run: nose2 -F -v -B --with-coverage --coverage TTS tests.zoo_tests.test_models.test_models_offset_2_step_3

2
.gitignore vendored
View File

@ -1,3 +1,5 @@
uv.lock
WadaSNR/
.idea/
*.pyc

28
.pre-commit-config.yaml
View File

@ -1,27 +1,19 @@
repos:
- repo: 'https://github.com/pre-commit/pre-commit-hooks'
rev: v2.3.0
- repo: "https://github.com/pre-commit/pre-commit-hooks"
rev: v5.0.0
hooks:
- id: check-json
files: "TTS/.models.json"
- id: check-yaml
- id: end-of-file-fixer
- id: trailing-whitespace
- repo: 'https://github.com/psf/black'
rev: 22.3.0
- repo: "https://github.com/psf/black"
rev: 24.2.0
hooks:
- id: black
language_version: python3
- repo: https://github.com/pycqa/isort
rev: 5.8.0
- repo: https://github.com/astral-sh/ruff-pre-commit
rev: v0.7.0
hooks:
- id: isort
name: isort (python)
- id: isort
name: isort (cython)
types: [cython]
- id: isort
name: isort (pyi)
types: [pyi]
- repo: https://github.com/pycqa/pylint
rev: v2.8.2
hooks:
- id: pylint
- id: ruff
args: [--fix, --exit-non-zero-on-fix]

599
.pylintrc
View File

@ -1,599 +0,0 @@
[MASTER]
# A comma-separated list of package or module names from where C extensions may
# be loaded. Extensions are loading into the active Python interpreter and may
# run arbitrary code.
extension-pkg-whitelist=
# Add files or directories to the blacklist. They should be base names, not
# paths.
ignore=CVS
# Add files or directories matching the regex patterns to the blacklist. The
# regex matches against base names, not paths.
ignore-patterns=
# Python code to execute, usually for sys.path manipulation such as
# pygtk.require().
#init-hook=
# Use multiple processes to speed up Pylint. Specifying 0 will auto-detect the
# number of processors available to use.
jobs=1
# Control the amount of potential inferred values when inferring a single
# object. This can help the performance when dealing with large functions or
# complex, nested conditions.
limit-inference-results=100
# List of plugins (as comma separated values of python modules names) to load,
# usually to register additional checkers.
load-plugins=
# Pickle collected data for later comparisons.
persistent=yes
# Specify a configuration file.
#rcfile=
# When enabled, pylint would attempt to guess common misconfiguration and emit
# user-friendly hints instead of false-positive error messages.
suggestion-mode=yes
# Allow loading of arbitrary C extensions. Extensions are imported into the
# active Python interpreter and may run arbitrary code.
unsafe-load-any-extension=no
[MESSAGES CONTROL]
# Only show warnings with the listed confidence levels. Leave empty to show
# all. Valid levels: HIGH, INFERENCE, INFERENCE_FAILURE, UNDEFINED.
confidence=
# Disable the message, report, category or checker with the given id(s). You
# can either give multiple identifiers separated by comma (,) or put this
# option multiple times (only on the command line, not in the configuration
# file where it should appear only once). You can also use "--disable=all" to
# disable everything first and then reenable specific checks. For example, if
# you want to run only the similarities checker, you can use "--disable=all
# --enable=similarities". If you want to run only the classes checker, but have
# no Warning level messages displayed, use "--disable=all --enable=classes
# --disable=W".
disable=missing-docstring,
too-many-public-methods,
too-many-lines,
bare-except,
## for avoiding weird p3.6 CI linter error
## TODO: see later if we can remove this
assigning-non-slot,
unsupported-assignment-operation,
## end
line-too-long,
fixme,
wrong-import-order,
ungrouped-imports,
wrong-import-position,
import-error,
invalid-name,
too-many-instance-attributes,
arguments-differ,
arguments-renamed,
no-name-in-module,
no-member,
unsubscriptable-object,
print-statement,
parameter-unpacking,
unpacking-in-except,
old-raise-syntax,
backtick,
long-suffix,
old-ne-operator,
old-octal-literal,
import-star-module-level,
non-ascii-bytes-literal,
raw-checker-failed,
bad-inline-option,
locally-disabled,
file-ignored,
suppressed-message,
useless-suppression,
deprecated-pragma,
use-symbolic-message-instead,
useless-object-inheritance,
too-few-public-methods,
too-many-branches,
too-many-arguments,
too-many-locals,
too-many-statements,
apply-builtin,
basestring-builtin,
buffer-builtin,
cmp-builtin,
coerce-builtin,
execfile-builtin,
file-builtin,
long-builtin,
raw_input-builtin,
reduce-builtin,
standarderror-builtin,
unicode-builtin,
xrange-builtin,
coerce-method,
delslice-method,
getslice-method,
setslice-method,
no-absolute-import,
old-division,
dict-iter-method,
dict-view-method,
next-method-called,
metaclass-assignment,
indexing-exception,
raising-string,
reload-builtin,
oct-method,
hex-method,
nonzero-method,
cmp-method,
input-builtin,
round-builtin,
intern-builtin,
unichr-builtin,
map-builtin-not-iterating,
zip-builtin-not-iterating,
range-builtin-not-iterating,
filter-builtin-not-iterating,
using-cmp-argument,
eq-without-hash,
div-method,
idiv-method,
rdiv-method,
exception-message-attribute,
invalid-str-codec,
sys-max-int,
bad-python3-import,
deprecated-string-function,
deprecated-str-translate-call,
deprecated-itertools-function,
deprecated-types-field,
next-method-defined,
dict-items-not-iterating,
dict-keys-not-iterating,
dict-values-not-iterating,
deprecated-operator-function,
deprecated-urllib-function,
xreadlines-attribute,
deprecated-sys-function,
exception-escape,
comprehension-escape,
duplicate-code,
not-callable,
import-outside-toplevel,
logging-fstring-interpolation,
logging-not-lazy
# Enable the message, report, category or checker with the given id(s). You can
# either give multiple identifier separated by comma (,) or put this option
# multiple time (only on the command line, not in the configuration file where
# it should appear only once). See also the "--disable" option for examples.
enable=c-extension-no-member
[REPORTS]
# Python expression which should return a note less than 10 (10 is the highest
# note). You have access to the variables errors warning, statement which
# respectively contain the number of errors / warnings messages and the total
# number of statements analyzed. This is used by the global evaluation report
# (RP0004).
evaluation=10.0 - ((float(5 * error + warning + refactor + convention) / statement) * 10)
# Template used to display messages. This is a python new-style format string
# used to format the message information. See doc for all details.
#msg-template=
# Set the output format. Available formats are text, parseable, colorized, json
# and msvs (visual studio). You can also give a reporter class, e.g.
# mypackage.mymodule.MyReporterClass.
output-format=text
# Tells whether to display a full report or only the messages.
reports=no
# Activate the evaluation score.
score=yes
[REFACTORING]
# Maximum number of nested blocks for function / method body
max-nested-blocks=5
# Complete name of functions that never returns. When checking for
# inconsistent-return-statements if a never returning function is called then
# it will be considered as an explicit return statement and no message will be
# printed.
never-returning-functions=sys.exit
[LOGGING]
# Format style used to check logging format string. `old` means using %
# formatting, while `new` is for `{}` formatting.
logging-format-style=old
# Logging modules to check that the string format arguments are in logging
# function parameter format.
logging-modules=logging
[SPELLING]
# Limits count of emitted suggestions for spelling mistakes.
max-spelling-suggestions=4
# Spelling dictionary name. Available dictionaries: none. To make it working
# install python-enchant package..
spelling-dict=
# List of comma separated words that should not be checked.
spelling-ignore-words=
# A path to a file that contains private dictionary; one word per line.
spelling-private-dict-file=
# Tells whether to store unknown words to indicated private dictionary in
# --spelling-private-dict-file option instead of raising a message.
spelling-store-unknown-words=no
[MISCELLANEOUS]
# List of note tags to take in consideration, separated by a comma.
notes=FIXME,
XXX,
TODO
[TYPECHECK]
# List of decorators that produce context managers, such as
# contextlib.contextmanager. Add to this list to register other decorators that
# produce valid context managers.
contextmanager-decorators=contextlib.contextmanager
# List of members which are set dynamically and missed by pylint inference
# system, and so shouldn't trigger E1101 when accessed. Python regular
# expressions are accepted.
generated-members=numpy.*,torch.*
# Tells whether missing members accessed in mixin class should be ignored. A
# mixin class is detected if its name ends with "mixin" (case insensitive).
ignore-mixin-members=yes
# Tells whether to warn about missing members when the owner of the attribute
# is inferred to be None.
ignore-none=yes
# This flag controls whether pylint should warn about no-member and similar
# checks whenever an opaque object is returned when inferring. The inference
# can return multiple potential results while evaluating a Python object, but
# some branches might not be evaluated, which results in partial inference. In
# that case, it might be useful to still emit no-member and other checks for
# the rest of the inferred objects.
ignore-on-opaque-inference=yes
# List of class names for which member attributes should not be checked (useful
# for classes with dynamically set attributes). This supports the use of
# qualified names.
ignored-classes=optparse.Values,thread._local,_thread._local
# List of module names for which member attributes should not be checked
# (useful for modules/projects where namespaces are manipulated during runtime
# and thus existing member attributes cannot be deduced by static analysis. It
# supports qualified module names, as well as Unix pattern matching.
ignored-modules=
# Show a hint with possible names when a member name was not found. The aspect
# of finding the hint is based on edit distance.
missing-member-hint=yes
# The minimum edit distance a name should have in order to be considered a
# similar match for a missing member name.
missing-member-hint-distance=1
# The total number of similar names that should be taken in consideration when
# showing a hint for a missing member.
missing-member-max-choices=1
[VARIABLES]
# List of additional names supposed to be defined in builtins. Remember that
# you should avoid defining new builtins when possible.
additional-builtins=
# Tells whether unused global variables should be treated as a violation.
allow-global-unused-variables=yes
# List of strings which can identify a callback function by name. A callback
# name must start or end with one of those strings.
callbacks=cb_,
_cb
# A regular expression matching the name of dummy variables (i.e. expected to
# not be used).
dummy-variables-rgx=_+$|(_[a-zA-Z0-9_]*[a-zA-Z0-9]+?$)|dummy|^ignored_|^unused_
# Argument names that match this expression will be ignored. Default to name
# with leading underscore.
ignored-argument-names=_.*|^ignored_|^unused_
# Tells whether we should check for unused import in __init__ files.
init-import=no
# List of qualified module names which can have objects that can redefine
# builtins.
redefining-builtins-modules=six.moves,past.builtins,future.builtins,builtins,io
[FORMAT]
# Expected format of line ending, e.g. empty (any line ending), LF or CRLF.
expected-line-ending-format=
# Regexp for a line that is allowed to be longer than the limit.
ignore-long-lines=^\s*(# )?<?https?://\S+>?$
# Number of spaces of indent required inside a hanging or continued line.
indent-after-paren=4
# String used as indentation unit. This is usually " " (4 spaces) or "\t" (1
# tab).
indent-string=' '
# Maximum number of characters on a single line.
max-line-length=120
# Maximum number of lines in a module.
max-module-lines=1000
# List of optional constructs for which whitespace checking is disabled. `dict-
# separator` is used to allow tabulation in dicts, etc.: {1 : 1,\n222: 2}.
# `trailing-comma` allows a space between comma and closing bracket: (a, ).
# `empty-line` allows space-only lines.
no-space-check=trailing-comma,
dict-separator
# Allow the body of a class to be on the same line as the declaration if body
# contains single statement.
single-line-class-stmt=no
# Allow the body of an if to be on the same line as the test if there is no
# else.
single-line-if-stmt=no
[SIMILARITIES]
# Ignore comments when computing similarities.
ignore-comments=yes
# Ignore docstrings when computing similarities.
ignore-docstrings=yes
# Ignore imports when computing similarities.
ignore-imports=no
# Minimum lines number of a similarity.
min-similarity-lines=4
[BASIC]
# Naming style matching correct argument names.
argument-naming-style=snake_case
# Regular expression matching correct argument names. Overrides argument-
# naming-style.
argument-rgx=[a-z_][a-z0-9_]{0,30}$
# Naming style matching correct attribute names.
attr-naming-style=snake_case
# Regular expression matching correct attribute names. Overrides attr-naming-
# style.
#attr-rgx=
# Bad variable names which should always be refused, separated by a comma.
bad-names=
# Naming style matching correct class attribute names.
class-attribute-naming-style=any
# Regular expression matching correct class attribute names. Overrides class-
# attribute-naming-style.
#class-attribute-rgx=
# Naming style matching correct class names.
class-naming-style=PascalCase
# Regular expression matching correct class names. Overrides class-naming-
# style.
#class-rgx=
# Naming style matching correct constant names.
const-naming-style=UPPER_CASE
# Regular expression matching correct constant names. Overrides const-naming-
# style.
#const-rgx=
# Minimum line length for functions/classes that require docstrings, shorter
# ones are exempt.
docstring-min-length=-1
# Naming style matching correct function names.
function-naming-style=snake_case
# Regular expression matching correct function names. Overrides function-
# naming-style.
#function-rgx=
# Good variable names which should always be accepted, separated by a comma.
good-names=i,
j,
k,
x,
ex,
Run,
_
# Include a hint for the correct naming format with invalid-name.
include-naming-hint=no
# Naming style matching correct inline iteration names.
inlinevar-naming-style=any
# Regular expression matching correct inline iteration names. Overrides
# inlinevar-naming-style.
#inlinevar-rgx=
# Naming style matching correct method names.
method-naming-style=snake_case
# Regular expression matching correct method names. Overrides method-naming-
# style.
#method-rgx=
# Naming style matching correct module names.
module-naming-style=snake_case
# Regular expression matching correct module names. Overrides module-naming-
# style.
#module-rgx=
# Colon-delimited sets of names that determine each other's naming style when
# the name regexes allow several styles.
name-group=
# Regular expression which should only match function or class names that do
# not require a docstring.
no-docstring-rgx=^_
# List of decorators that produce properties, such as abc.abstractproperty. Add
# to this list to register other decorators that produce valid properties.
# These decorators are taken in consideration only for invalid-name.
property-classes=abc.abstractproperty
# Naming style matching correct variable names.
variable-naming-style=snake_case
# Regular expression matching correct variable names. Overrides variable-
# naming-style.
variable-rgx=[a-z_][a-z0-9_]{0,30}$
[STRING]
# This flag controls whether the implicit-str-concat-in-sequence should
# generate a warning on implicit string concatenation in sequences defined over
# several lines.
check-str-concat-over-line-jumps=no
[IMPORTS]
# Allow wildcard imports from modules that define __all__.
allow-wildcard-with-all=no
# Analyse import fallback blocks. This can be used to support both Python 2 and
# 3 compatible code, which means that the block might have code that exists
# only in one or another interpreter, leading to false positives when analysed.
analyse-fallback-blocks=no
# Deprecated modules which should not be used, separated by a comma.
deprecated-modules=optparse,tkinter.tix
# Create a graph of external dependencies in the given file (report RP0402 must
# not be disabled).
ext-import-graph=
# Create a graph of every (i.e. internal and external) dependencies in the
# given file (report RP0402 must not be disabled).
import-graph=
# Create a graph of internal dependencies in the given file (report RP0402 must
# not be disabled).
int-import-graph=
# Force import order to recognize a module as part of the standard
# compatibility libraries.
known-standard-library=
# Force import order to recognize a module as part of a third party library.
known-third-party=enchant
[CLASSES]
# List of method names used to declare (i.e. assign) instance attributes.
defining-attr-methods=__init__,
__new__,
setUp
# List of member names, which should be excluded from the protected access
# warning.
exclude-protected=_asdict,
_fields,
_replace,
_source,
_make
# List of valid names for the first argument in a class method.
valid-classmethod-first-arg=cls
# List of valid names for the first argument in a metaclass class method.
valid-metaclass-classmethod-first-arg=cls
[DESIGN]
# Maximum number of arguments for function / method.
max-args=5
# Maximum number of attributes for a class (see R0902).
max-attributes=7
# Maximum number of boolean expressions in an if statement.
max-bool-expr=5
# Maximum number of branch for function / method body.
max-branches=12
# Maximum number of locals for function / method body.
max-locals=15
# Maximum number of parents for a class (see R0901).
max-parents=15
# Maximum number of public methods for a class (see R0904).
max-public-methods=20
# Maximum number of return / yield for function / method body.
max-returns=6
# Maximum number of statements in function / method body.
max-statements=50
# Minimum number of public methods for a class (see R0903).
min-public-methods=2
[EXCEPTIONS]
# Exceptions that will emit a warning when being caught. Defaults to
# "BaseException, Exception".
overgeneral-exceptions=BaseException,
Exception

14
.readthedocs.yml
View File

@ -9,13 +9,13 @@ version: 2
build:
os: ubuntu-22.04
tools:
python: "3.11"
# Optionally set the version of Python and requirements required to build your docs
python:
install:
- requirements: docs/requirements.txt
- requirements: requirements.txt
python: "3.12"
commands:
- asdf plugin add uv
- asdf install uv latest
- asdf global uv latest
- uv sync --group docs
- uv run -m sphinx -T -b html -d docs/_build/doctrees -D language=en docs/source $READTHEDOCS_OUTPUT/html
# Build documentation in the docs/ directory with Sphinx
sphinx:

4
CITATION.cff
View File

@ -10,8 +10,8 @@ authors:
version: 1.4
doi: 10.5281/zenodo.6334862
license: "MPL-2.0"
url: "https://www.coqui.ai"
repository-code: "https://github.com/coqui-ai/TTS"
url: "https://github.com/idiap/coqui-ai-TTS"
repository-code: "https://github.com/idiap/coqui-ai-TTS"
keywords:
- machine learning
- deep learning

75
CODE_OWNERS.rst
View File

@ -1,75 +0,0 @@
TTS code owners / governance system
==========================================
TTS is run under a governance system inspired (and partially copied from) by the `Mozilla module ownership system <https://www.mozilla.org/about/governance/policies/module-ownership/>`_. The project is roughly divided into modules, and each module has its owners, which are responsible for reviewing pull requests and deciding on technical direction for their modules. Module ownership authority is given to people who have worked extensively on areas of the project.
Module owners also have the authority of naming other module owners or appointing module peers, which are people with authority to review pull requests in that module. They can also sub-divide their module into sub-modules with their owners.
Module owners are not tyrants. They are chartered to make decisions with input from the community and in the best interest of the community. Module owners are not required to make code changes or additions solely because the community wants them to do so. (Like anyone else, the module owners may write code because they want to, because their employers want them to, because the community wants them to, or for some other reason.) Module owners do need to pay attention to patches submitted to that module. However “pay attention” does not mean agreeing to every patch. Some patches may not make sense for the WebThings project; some may be poorly implemented. Module owners have the authority to decline a patch; this is a necessary part of the role. We ask the module owners to describe in the relevant issue their reasons for wanting changes to a patch, for declining it altogether, or for postponing review for some period. We don’t ask or expect them to rewrite patches to make them acceptable. Similarly, module owners may need to delay review of a promising patch due to an upcoming deadline. For example, a patch may be of interest, but not for the next milestone. In such a case it may make sense for the module owner to postpone review of a patch until after matters needed for a milestone have been finalized. Again, we expect this to be described in the relevant issue. And of course, it shouldn’t go on very often or for very long or escalation and review is likely.
The work of the various module owners and peers is overseen by the global owners, which are responsible for making final decisions in case there's conflict between owners as well as set the direction for the project as a whole.
This file describes module owners who are active on the project and which parts of the code they have expertise on (and interest in). If you're making changes to the code and are wondering who's an appropriate person to talk to, this list will tell you who to ping.
There's overlap in the areas of expertise of each owner, and in particular when looking at which files are covered by each area, there is a lot of overlap. Don't worry about getting it exactly right when requesting review, any code owner will be happy to redirect the request to a more appropriate person.
Global owners
----------------
These are people who have worked on the project extensively and are familiar with all or most parts of it. Their expertise and review guidance is trusted by other code owners to cover their own areas of expertise. In case of conflicting opinions from other owners, global owners will make a final decision.
- Eren Gölge (@erogol)
- Reuben Morais (@reuben)
Training, feeding
-----------------
- Eren Gölge (@erogol)
Model exporting
---------------
- Eren Gölge (@erogol)
Multi-Speaker TTS
-----------------
- Eren Gölge (@erogol)
- Edresson Casanova (@edresson)
TTS
---
- Eren Gölge (@erogol)
Vocoders
--------
- Eren Gölge (@erogol)
Speaker Encoder
---------------
- Eren Gölge (@erogol)
Testing & CI
------------
- Eren Gölge (@erogol)
- Reuben Morais (@reuben)
Python bindings
---------------
- Eren Gölge (@erogol)
- Reuben Morais (@reuben)
Documentation
-------------
- Eren Gölge (@erogol)
Third party bindings
--------------------
Owned by the author.

87
CONTRIBUTING.md
View File

@ -2,7 +2,7 @@
Welcome to the 🐸TTS!
This repository is governed by [the Contributor Covenant Code of Conduct](https://github.com/coqui-ai/TTS/blob/main/CODE_OF_CONDUCT.md).
This repository is governed by [the Contributor Covenant Code of Conduct](https://github.com/idiap/coqui-ai-TTS/blob/main/CODE_OF_CONDUCT.md).
## Where to start.
We welcome everyone who likes to contribute to 🐸TTS.
@ -11,30 +11,25 @@ You can contribute not only with code but with bug reports, comments, questions,
If you like to contribute code, squash a bug but if you don't know where to start, here are some pointers.
- [Development Road Map](https://github.com/coqui-ai/TTS/issues/378)
You can pick something out of our road map. We keep the progess of the project in this simple issue thread. It has new model proposals or developmental updates etc.
- [Github Issues Tracker](https://github.com/coqui-ai/TTS/issues)
- [Github Issues Tracker](https://github.com/idiap/coqui-ai-TTS/issues)
This is a place to find feature requests, bugs.
Issues with the ```good first issue``` tag are good place for beginners to take on.
- ✨**PR**✨ [pages](https://github.com/coqui-ai/TTS/pulls) with the ```🚀new version``` tag.
We list all the target improvements for the next version. You can pick one of them and start contributing.
Issues with the ```good first issue``` tag are good place for beginners to
take on. Issues tagged with `help wanted` are suited for more experienced
outside contributors.
- Also feel free to suggest new features, ideas and models. We're always open for new things.
## Call for sharing language models
## Call for sharing pretrained models
If possible, please consider sharing your pre-trained models in any language (if the licences allow for you to do so). We will include them in our model catalogue for public use and give the proper attribution, whether it be your name, company, website or any other source specified.
This model can be shared in two ways:
1. Share the model files with us and we serve them with the next 🐸 TTS release.
2. Upload your models on GDrive and share the link.
Models are served under `.models.json` file and any model is available under TTS CLI or Server end points.
Models are served under `.models.json` file and any model is available under TTS
CLI and Python API end points.
Either way you choose, please make sure you send the models [here](https://github.com/coqui-ai/TTS/discussions/930).
@ -44,29 +39,37 @@ If you have a new feature, a model to implement, or a bug to squash, go ahead an
Please use the following steps to send a ✨**PR**✨.
Let us know if you encounter a problem along the way.
The following steps are tested on an Ubuntu system.
The following steps are tested on an Ubuntu system and require
[uv](https://docs.astral.sh/uv/) for virtual environment management. Choose your
preferred [installation
method](https://docs.astral.sh/uv/getting-started/installation/), e.g. the
standalone installer:
1. Fork 🐸TTS[https://github.com/coqui-ai/TTS] by clicking the fork button at the top right corner of the project page.
```bash
curl -LsSf https://astral.sh/uv/install.sh | sh
```
1. Fork 🐸TTS[https://github.com/idiap/coqui-ai-TTS] by clicking the fork button at the top right corner of the project page.
2. Clone 🐸TTS and add the main repo as a new remote named ```upstream```.
```bash
$ git clone git@github.com:<your Github name>/TTS.git
$ cd TTS
$ git remote add upstream https://github.com/coqui-ai/TTS.git
git clone git@github.com:<your Github name>/coqui-ai-TTS.git
cd coqui-ai-TTS
git remote add upstream https://github.com/idiap/coqui-ai-TTS.git
```
3. Install 🐸TTS for development.
```bash
$ make system-deps # intended to be used on Ubuntu (Debian). Let us know if you have a different OS.
$ make install
make system-deps # intended to be used on Ubuntu (Debian). Let us know if you have a different OS.
make install_dev
```
4. Create a new branch with an informative name for your goal.
```bash
$ git checkout -b an_informative_name_for_my_branch
git checkout -b an_informative_name_for_my_branch
```
5. Implement your changes on your new branch.
@ -75,39 +78,42 @@ The following steps are tested on an Ubuntu system.
7. Add your tests to our test suite under ```tests``` folder. It is important to show that your code works, edge cases are considered, and inform others about the intended use.
8. Run the tests to see how your updates work with the rest of the project. You can repeat this step multiple times as you implement your changes to make sure you are on the right direction.
8. Run the tests to see how your updates work with the rest of the project. You
can repeat this step multiple times as you implement your changes to make
sure you are on the right direction. **NB: running all tests takes a long time,
it is better to leave this to the CI.**
```bash
$ make test # stop at the first error
$ make test_all # run all the tests, report all the errors
uv run make test # stop at the first error
uv run make test_all # run all the tests, report all the errors
```
9. Format your code. We use ```black``` for code and ```isort``` for ```import``` formatting.
9. Format your code. We use ```black``` for code formatting.
```bash
$ make style
make style
```
10. Run the linter and correct the issues raised. We use ```pylint``` for linting. It helps to enforce a coding standard, offers simple refactoring suggestions.
10. Run the linter and correct the issues raised. We use ```ruff``` for linting. It helps to enforce a coding standard, offers simple refactoring suggestions.
```bash
$ make lint
make lint
```
11. When things are good, add new files and commit your changes.
```bash
$ git add my_file1.py my_file2.py ...
$ git commit
git add my_file1.py my_file2.py ...
git commit
```
It's a good practice to regularly sync your local copy of the project with the upstream code to keep up with the recent updates.
```bash
$ git fetch upstream
$ git rebase upstream/master
git fetch upstream
git rebase upstream/main
# or for the development version
$ git rebase upstream/dev
git rebase upstream/dev
```
12. Send a PR to ```dev``` branch.
@ -115,7 +121,7 @@ The following steps are tested on an Ubuntu system.
Push your branch to your fork.
```bash
$ git push -u origin an_informative_name_for_my_branch
git push -u origin an_informative_name_for_my_branch
```
Then go to your fork's Github page and click on 'Pull request' to send your ✨**PR**✨.
@ -124,7 +130,8 @@ The following steps are tested on an Ubuntu system.
13. Let's discuss until it is perfect. 💪
We might ask you for certain changes that would appear in the ✨**PR**✨'s page under 🐸TTS[https://github.com/coqui-ai/TTS/pulls].
We might ask you for certain changes that would appear in the
[Github ✨**PR**✨'s page](https://github.com/idiap/coqui-ai-TTS/pulls).
14. Once things look perfect, We merge it to the ```dev``` branch and make it ready for the next version.
@ -132,14 +139,14 @@ The following steps are tested on an Ubuntu system.
If you prefer working within a Docker container as your development environment, you can do the following:
1. Fork 🐸TTS[https://github.com/coqui-ai/TTS] by clicking the fork button at the top right corner of the project page.
1. Fork the 🐸TTS [Github repository](https://github.com/idiap/coqui-ai-TTS) by clicking the fork button at the top right corner of the page.
2. Clone 🐸TTS and add the main repo as a new remote named ```upsteam```.
2. Clone 🐸TTS and add the main repo as a new remote named ```upstream```.
```bash
$ git clone git@github.com:<your Github name>/TTS.git
$ cd TTS
$ git remote add upstream https://github.com/coqui-ai/TTS.git
git clone git@github.com:<your Github name>/coqui-ai-TTS.git
cd coqui-ai-TTS
git remote add upstream https://github.com/idiap/coqui-ai-TTS.git
```
3. Build the Docker Image as your development environment (it installs all of the dependencies for you):

18
Dockerfile
View File

@ -1,8 +1,20 @@
ARG BASE=nvidia/cuda:11.8.0-base-ubuntu22.04
FROM ${BASE}
RUN apt-get update && apt-get upgrade -y
RUN apt-get install -y --no-install-recommends gcc g++ make python3 python3-dev python3-pip python3-venv python3-wheel espeak-ng libsndfile1-dev && rm -rf /var/lib/apt/lists/*
RUN apt-get update && \
apt-get upgrade -y
RUN apt-get install -y --no-install-recommends \
gcc g++ make python3 python3-dev python3-pip \
python3-venv python3-wheel espeak-ng \
libsndfile1-dev libc-dev curl && \
rm -rf /var/lib/apt/lists/*
# Install Rust compiler (to build sudachipy for Mac)
RUN curl --proto '=https' --tlsv1.2 -sSf "https://sh.rustup.rs" | sh -s -- -y
ENV PATH="/root/.cargo/bin:${PATH}"
RUN pip3 install -U pip setuptools wheel
RUN pip3 install -U "spacy[ja]<3.8"
RUN pip3 install llvmlite --ignore-installed
# Install Dependencies:
@ -13,7 +25,7 @@ RUN rm -rf /root/.cache/pip
WORKDIR /root
COPY . /root
RUN make install
RUN pip3 install -e .[all]
ENTRYPOINT ["tts"]
CMD ["--help"]

15
MANIFEST.in
View File

@ -1,15 +0,0 @@
include README.md
include LICENSE.txt
include requirements.*.txt
include *.cff
include requirements.txt
include TTS/VERSION
recursive-include TTS *.json
recursive-include TTS *.html
recursive-include TTS *.png
recursive-include TTS *.md
recursive-include TTS *.py
recursive-include TTS *.pyx
recursive-include images *.png
recursive-exclude tests *
prune tests*

64
Makefile
View File

@ -1,5 +1,5 @@
.DEFAULT_GOAL := help
.PHONY: test system-deps dev-deps deps style lint install help docs
.PHONY: test system-deps style lint install install_dev help docs
help:
@grep -E '^[a-zA-Z_-]+:.*?## .*$$' $(MAKEFILE_LIST) | sort | awk 'BEGIN {FS = ":.*?## "}; {printf "\033[36m%-30s\033[0m %s\n", $$1, $$2}'
@ -11,68 +11,60 @@ test_all: ## run tests and don't stop on an error.
./run_bash_tests.sh
test: ## run tests.
nose2 -F -v -B --with-coverage --coverage TTS tests
coverage run -m nose2 -F -v -B tests
test_vocoder: ## run vocoder tests.
nose2 -F -v -B --with-coverage --coverage TTS tests.vocoder_tests
coverage run -m nose2 -F -v -B tests.vocoder_tests
test_tts: ## run tts tests.
nose2 -F -v -B --with-coverage --coverage TTS tests.tts_tests
coverage run -m nose2 -F -v -B tests.tts_tests
test_tts2: ## run tts tests.
nose2 -F -v -B --with-coverage --coverage TTS tests.tts_tests2
coverage run -m nose2 -F -v -B tests.tts_tests2
test_xtts:
nose2 -F -v -B --with-coverage --coverage TTS tests.xtts_tests
coverage run -m nose2 -F -v -B tests.xtts_tests
test_aux: ## run aux tests.
nose2 -F -v -B --with-coverage --coverage TTS tests.aux_tests
coverage run -m nose2 -F -v -B tests.aux_tests
./run_bash_tests.sh
test_zoo: ## run zoo tests.
nose2 -F -v -B --with-coverage --coverage TTS tests.zoo_tests
test_zoo0: ## run zoo tests.
coverage run -m nose2 -F -v -B tests.zoo_tests.test_models.test_models_offset_0_step_3 \
tests.zoo_tests.test_models.test_voice_conversion
test_zoo1: ## run zoo tests.
coverage run -m nose2 -F -v -B tests.zoo_tests.test_models.test_models_offset_1_step_3
test_zoo2: ## run zoo tests.
coverage run -m nose2 -F -v -B tests.zoo_tests.test_models.test_models_offset_2_step_3
inference_tests: ## run inference tests.
nose2 -F -v -B --with-coverage --coverage TTS tests.inference_tests
coverage run -m nose2 -F -v -B tests.inference_tests
data_tests: ## run data tests.
nose2 -F -v -B --with-coverage --coverage TTS tests.data_tests
coverage run -m nose2 -F -v -B tests.data_tests
test_text: ## run text tests.
nose2 -F -v -B --with-coverage --coverage TTS tests.text_tests
coverage run -m nose2 -F -v -B tests.text_tests
test_failed: ## only run tests failed the last time.
nose2 -F -v -B --with-coverage --coverage TTS tests
coverage run -m nose2 -F -v -B tests
style: ## update code style.
black ${target_dirs}
isort ${target_dirs}
uv run --only-dev black ${target_dirs}
lint: ## run pylint linter.
pylint ${target_dirs}
black ${target_dirs} --check
isort ${target_dirs} --check-only
lint: ## run linters.
uv run --only-dev ruff check ${target_dirs}
uv run --only-dev black ${target_dirs} --check
system-deps: ## install linux system deps
sudo apt-get install -y libsndfile1-dev
dev-deps: ## install development deps
pip install -r requirements.dev.txt
install: ## install 🐸 TTS
uv sync --all-extras
doc-deps: ## install docs dependencies
pip install -r docs/requirements.txt
build-docs: ## build the docs
cd docs && make clean && make build
hub-deps: ## install deps for torch hub use
pip install -r requirements.hub.txt
deps: ## install 🐸 requirements.
pip install -r requirements.txt
install: ## install 🐸 TTS for development.
pip install -e .[all]
install_dev: ## install 🐸 TTS for development.
uv sync --all-extras
uv run pre-commit install
docs: ## build the docs
$(MAKE) -C docs clean && $(MAKE) -C docs html
uv run --group docs $(MAKE) -C docs clean && uv run --group docs $(MAKE) -C docs html

434
README.md
View File

@ -1,177 +1,173 @@
## 🐸Coqui.ai News
- 📣 ⓍTTSv2 is here with 16 languages and better performance across the board.
- 📣 ⓍTTS fine-tuning code is out. Check the [example recipes](https://github.com/coqui-ai/TTS/tree/dev/recipes/ljspeech).
- 📣 ⓍTTS can now stream with <200ms latency.
- 📣 ⓍTTS, our production TTS model that can speak 13 languages, is released [Blog Post](https://coqui.ai/blog/tts/open_xtts), [Demo](https://huggingface.co/spaces/coqui/xtts), [Docs](https://tts.readthedocs.io/en/dev/models/xtts.html)
- 📣 [🐶Bark](https://github.com/suno-ai/bark) is now available for inference with unconstrained voice cloning. [Docs](https://tts.readthedocs.io/en/dev/models/bark.html)
- 📣 You can use [~1100 Fairseq models](https://github.com/facebookresearch/fairseq/tree/main/examples/mms) with 🐸TTS.
- 📣 🐸TTS now supports 🐢Tortoise with faster inference. [Docs](https://tts.readthedocs.io/en/dev/models/tortoise.html)
<div align="center">
<img src="https://static.scarf.sh/a.png?x-pxid=cf317fe7-2188-4721-bc01-124bb5d5dbb2" />
## <img src="https://raw.githubusercontent.com/coqui-ai/TTS/main/images/coqui-log-green-TTS.png" height="56"/>
# <img src="https://raw.githubusercontent.com/idiap/coqui-ai-TTS/main/images/coqui-log-green-TTS.png" height="56"/>
**🐸TTS is a library for advanced Text-to-Speech generation.**
**🐸 Coqui TTS is a library for advanced Text-to-Speech generation.**
🚀 Pretrained models in +1100 languages.
🛠️ Tools for training new models and fine-tuning existing models in any language.
📚 Utilities for dataset analysis and curation.
______________________________________________________________________
[![Discord](https://img.shields.io/discord/1037326658807533628?color=%239B59B6&label=chat%20on%20discord)](https://discord.gg/5eXr5seRrv)
[![PyPI - Python Version](https://img.shields.io/pypi/pyversions/coqui-tts)](https://pypi.org/project/coqui-tts/)
[![License](<https://img.shields.io/badge/License-MPL%202.0-brightgreen.svg>)](https://opensource.org/licenses/MPL-2.0)
[![PyPI version](https://badge.fury.io/py/TTS.svg)](https://badge.fury.io/py/TTS)
[![Covenant](https://camo.githubusercontent.com/7d620efaa3eac1c5b060ece5d6aacfcc8b81a74a04d05cd0398689c01c4463bb/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f436f6e7472696275746f72253230436f76656e616e742d76322e3025323061646f707465642d6666363962342e737667)](https://github.com/coqui-ai/TTS/blob/master/CODE_OF_CONDUCT.md)
[![Downloads](https://pepy.tech/badge/tts)](https://pepy.tech/project/tts)
[![PyPI version](https://badge.fury.io/py/coqui-tts.svg)](https://pypi.org/project/coqui-tts/)
[![Downloads](https://pepy.tech/badge/coqui-tts)](https://pepy.tech/project/coqui-tts)
[![DOI](https://zenodo.org/badge/265612440.svg)](https://zenodo.org/badge/latestdoi/265612440)
![GithubActions](https://github.com/coqui-ai/TTS/actions/workflows/aux_tests.yml/badge.svg)
![GithubActions](https://github.com/coqui-ai/TTS/actions/workflows/data_tests.yml/badge.svg)
![GithubActions](https://github.com/coqui-ai/TTS/actions/workflows/docker.yaml/badge.svg)
![GithubActions](https://github.com/coqui-ai/TTS/actions/workflows/inference_tests.yml/badge.svg)
![GithubActions](https://github.com/coqui-ai/TTS/actions/workflows/style_check.yml/badge.svg)
![GithubActions](https://github.com/coqui-ai/TTS/actions/workflows/text_tests.yml/badge.svg)
![GithubActions](https://github.com/coqui-ai/TTS/actions/workflows/tts_tests.yml/badge.svg)
![GithubActions](https://github.com/coqui-ai/TTS/actions/workflows/vocoder_tests.yml/badge.svg)
![GithubActions](https://github.com/coqui-ai/TTS/actions/workflows/zoo_tests0.yml/badge.svg)
![GithubActions](https://github.com/coqui-ai/TTS/actions/workflows/zoo_tests1.yml/badge.svg)
![GithubActions](https://github.com/coqui-ai/TTS/actions/workflows/zoo_tests2.yml/badge.svg)
[![Docs](<https://readthedocs.org/projects/tts/badge/?version=latest&style=plastic>)](https://tts.readthedocs.io/en/latest/)
[![GithubActions](https://github.com/idiap/coqui-ai-TTS/actions/workflows/tests.yml/badge.svg)](https://github.com/idiap/coqui-ai-TTS/actions/workflows/tests.yml)
[![GithubActions](https://github.com/idiap/coqui-ai-TTS/actions/workflows/docker.yaml/badge.svg)](https://github.com/idiap/coqui-ai-TTS/actions/workflows/docker.yaml)
[![GithubActions](https://github.com/idiap/coqui-ai-TTS/actions/workflows/style_check.yml/badge.svg)](https://github.com/idiap/coqui-ai-TTS/actions/workflows/style_check.yml)
[![Docs](<https://readthedocs.org/projects/coqui-tts/badge/?version=latest&style=plastic>)](https://coqui-tts.readthedocs.io/en/latest/)
</div>
______________________________________________________________________
## 📣 News
- **Fork of the [original, unmaintained repository](https://github.com/coqui-ai/TTS). New PyPI package: [coqui-tts](https://pypi.org/project/coqui-tts)**
- 0.25.0: [OpenVoice](https://github.com/myshell-ai/OpenVoice) models now available for voice conversion.
- 0.24.2: Prebuilt wheels are now also published for Mac and Windows (in addition to Linux as before) for easier installation across platforms.
- 0.20.0: XTTSv2 is here with 17 languages and better performance across the board. XTTS can stream with <200ms latency.
- 0.19.0: XTTS fine-tuning code is out. Check the [example recipes](https://github.com/idiap/coqui-ai-TTS/tree/dev/recipes/ljspeech).
- 0.14.1: You can use [Fairseq models in ~1100 languages](https://github.com/facebookresearch/fairseq/tree/main/examples/mms) with 🐸TTS.
## 💬 Where to ask questions
Please use our dedicated channels for questions and discussion. Help is much more valuable if it's shared publicly so that more people can benefit from it.
| Type | Platforms |
| ------------------------------- | --------------------------------------- |
| 🚨 **Bug Reports** | [GitHub Issue Tracker] |
| 🎁 **Feature Requests & Ideas** | [GitHub Issue Tracker] |
| -------------------------------------------- | ----------------------------------- |
| 🚨 **Bug Reports, Feature Requests & Ideas** | [GitHub Issue Tracker] |
| 👩‍💻 **Usage Questions** | [GitHub Discussions] |
| 🗯 **General Discussion** | [GitHub Discussions] or [Discord] |
[github issue tracker]: https://github.com/coqui-ai/tts/issues
[github discussions]: https://github.com/coqui-ai/TTS/discussions
[github issue tracker]: https://github.com/idiap/coqui-ai-TTS/issues
[github discussions]: https://github.com/idiap/coqui-ai-TTS/discussions
[discord]: https://discord.gg/5eXr5seRrv
[Tutorials and Examples]: https://github.com/coqui-ai/TTS/wiki/TTS-Notebooks-and-Tutorials
The [issues](https://github.com/coqui-ai/TTS/issues) and
[discussions](https://github.com/coqui-ai/TTS/discussions) in the original
repository are also still a useful source of information.
## 🔗 Links and Resources
| Type | Links |
| ------------------------------- | --------------------------------------- |
| 💼 **Documentation** | [ReadTheDocs](https://tts.readthedocs.io/en/latest/)
| 💾 **Installation** | [TTS/README.md](https://github.com/coqui-ai/TTS/tree/dev#installation)|
| 👩‍💻 **Contributing** | [CONTRIBUTING.md](https://github.com/coqui-ai/TTS/blob/main/CONTRIBUTING.md)|
| 📌 **Road Map** | [Main Development Plans](https://github.com/coqui-ai/TTS/issues/378)
| 🚀 **Released Models** | [TTS Releases](https://github.com/coqui-ai/TTS/releases) and [Experimental Models](https://github.com/coqui-ai/TTS/wiki/Experimental-Released-Models)|
| 📰 **Papers** | [TTS Papers](https://github.com/erogol/TTS-papers)|
## 🥇 TTS Performance
<p align="center"><img src="https://raw.githubusercontent.com/coqui-ai/TTS/main/images/TTS-performance.png" width="800" /></p>
Underlined "TTS*" and "Judy*" are **internal** 🐸TTS models that are not released open-source. They are here to show the potential. Models prefixed with a dot (.Jofish .Abe and .Janice) are real human voices.
| 💼 **Documentation** | [ReadTheDocs](https://coqui-tts.readthedocs.io/en/latest/)
| 💾 **Installation** | [TTS/README.md](https://github.com/idiap/coqui-ai-TTS/tree/dev#installation)|
| 👩‍💻 **Contributing** | [CONTRIBUTING.md](https://github.com/idiap/coqui-ai-TTS/blob/main/CONTRIBUTING.md)|
| 🚀 **Released Models** | [Standard models](https://github.com/idiap/coqui-ai-TTS/blob/dev/TTS/.models.json) and [Fairseq models in ~1100 languages](https://github.com/idiap/coqui-ai-TTS#example-text-to-speech-using-fairseq-models-in-1100-languages-)|
## Features
- High-performance Deep Learning models for Text2Speech tasks.
- Text2Spec models (Tacotron, Tacotron2, Glow-TTS, SpeedySpeech).
- Speaker Encoder to compute speaker embeddings efficiently.
- Vocoder models (MelGAN, Multiband-MelGAN, GAN-TTS, ParallelWaveGAN, WaveGrad, WaveRNN)
- Fast and efficient model training.
- Detailed training logs on the terminal and Tensorboard.
- Support for Multi-speaker TTS.
- Efficient, flexible, lightweight but feature complete `Trainer API`.
- High-performance text-to-speech and voice conversion models, see list below.
- Fast and efficient model training with detailed training logs on the terminal and Tensorboard.
- Support for multi-speaker and multilingual TTS.
- Released and ready-to-use models.
- Tools to curate Text2Speech datasets under```dataset_analysis```.
- Utilities to use and test your models.
- Tools to curate TTS datasets under ```dataset_analysis/```.
- Command line and Python APIs to use and test your models.
- Modular (but not too much) code base enabling easy implementation of new ideas.
## Model Implementations
### Spectrogram models
- Tacotron: [paper](https://arxiv.org/abs/1703.10135)
- Tacotron2: [paper](https://arxiv.org/abs/1712.05884)
- Glow-TTS: [paper](https://arxiv.org/abs/2005.11129)
- Speedy-Speech: [paper](https://arxiv.org/abs/2008.03802)
- Align-TTS: [paper](https://arxiv.org/abs/2003.01950)
- FastPitch: [paper](https://arxiv.org/pdf/2006.06873.pdf)
- FastSpeech: [paper](https://arxiv.org/abs/1905.09263)
- FastSpeech2: [paper](https://arxiv.org/abs/2006.04558)
- SC-GlowTTS: [paper](https://arxiv.org/abs/2104.05557)
- Capacitron: [paper](https://arxiv.org/abs/1906.03402)
- OverFlow: [paper](https://arxiv.org/abs/2211.06892)
- Neural HMM TTS: [paper](https://arxiv.org/abs/2108.13320)
- Delightful TTS: [paper](https://arxiv.org/abs/2110.12612)
- [Tacotron](https://arxiv.org/abs/1703.10135), [Tacotron2](https://arxiv.org/abs/1712.05884)
- [Glow-TTS](https://arxiv.org/abs/2005.11129), [SC-GlowTTS](https://arxiv.org/abs/2104.05557)
- [Speedy-Speech](https://arxiv.org/abs/2008.03802)
- [Align-TTS](https://arxiv.org/abs/2003.01950)
- [FastPitch](https://arxiv.org/pdf/2006.06873.pdf)
- [FastSpeech](https://arxiv.org/abs/1905.09263), [FastSpeech2](https://arxiv.org/abs/2006.04558)
- [Capacitron](https://arxiv.org/abs/1906.03402)
- [OverFlow](https://arxiv.org/abs/2211.06892)
- [Neural HMM TTS](https://arxiv.org/abs/2108.13320)
- [Delightful TTS](https://arxiv.org/abs/2110.12612)
### End-to-End Models
- ⓍTTS: [blog](https://coqui.ai/blog/tts/open_xtts)
- VITS: [paper](https://arxiv.org/pdf/2106.06103)
- 🐸 YourTTS: [paper](https://arxiv.org/abs/2112.02418)
- 🐢 Tortoise: [orig. repo](https://github.com/neonbjb/tortoise-tts)
- 🐶 Bark: [orig. repo](https://github.com/suno-ai/bark)
### Attention Methods
- Guided Attention: [paper](https://arxiv.org/abs/1710.08969)
- Forward Backward Decoding: [paper](https://arxiv.org/abs/1907.09006)
- Graves Attention: [paper](https://arxiv.org/abs/1910.10288)
- Double Decoder Consistency: [blog](https://erogol.com/solving-attention-problems-of-tts-models-with-double-decoder-consistency/)
- Dynamic Convolutional Attention: [paper](https://arxiv.org/pdf/1910.10288.pdf)
- Alignment Network: [paper](https://arxiv.org/abs/2108.10447)
### Speaker Encoder
- GE2E: [paper](https://arxiv.org/abs/1710.10467)
- Angular Loss: [paper](https://arxiv.org/pdf/2003.11982.pdf)
- [XTTS](https://arxiv.org/abs/2406.04904)
- [VITS](https://arxiv.org/pdf/2106.06103)
- 🐸[YourTTS](https://arxiv.org/abs/2112.02418)
- 🐢[Tortoise](https://github.com/neonbjb/tortoise-tts)
- 🐶[Bark](https://github.com/suno-ai/bark)
### Vocoders
- MelGAN: [paper](https://arxiv.org/abs/1910.06711)
- MultiBandMelGAN: [paper](https://arxiv.org/abs/2005.05106)
- ParallelWaveGAN: [paper](https://arxiv.org/abs/1910.11480)
- GAN-TTS discriminators: [paper](https://arxiv.org/abs/1909.11646)
- WaveRNN: [origin](https://github.com/fatchord/WaveRNN/)
- WaveGrad: [paper](https://arxiv.org/abs/2009.00713)
- HiFiGAN: [paper](https://arxiv.org/abs/2010.05646)
- UnivNet: [paper](https://arxiv.org/abs/2106.07889)
- [MelGAN](https://arxiv.org/abs/1910.06711)
- [MultiBandMelGAN](https://arxiv.org/abs/2005.05106)
- [ParallelWaveGAN](https://arxiv.org/abs/1910.11480)
- [GAN-TTS discriminators](https://arxiv.org/abs/1909.11646)
- [WaveRNN](https://github.com/fatchord/WaveRNN/)
- [WaveGrad](https://arxiv.org/abs/2009.00713)
- [HiFiGAN](https://arxiv.org/abs/2010.05646)
- [UnivNet](https://arxiv.org/abs/2106.07889)
### Voice Conversion
- FreeVC: [paper](https://arxiv.org/abs/2210.15418)
- [FreeVC](https://arxiv.org/abs/2210.15418)
- [OpenVoice](https://arxiv.org/abs/2312.01479)
### Others
- Attention methods: [Guided Attention](https://arxiv.org/abs/1710.08969),
[Forward Backward Decoding](https://arxiv.org/abs/1907.09006),
[Graves Attention](https://arxiv.org/abs/1910.10288),
[Double Decoder Consistency](https://erogol.com/solving-attention-problems-of-tts-models-with-double-decoder-consistency/),
[Dynamic Convolutional Attention](https://arxiv.org/pdf/1910.10288.pdf),
[Alignment Network](https://arxiv.org/abs/2108.10447)
- Speaker encoders: [GE2E](https://arxiv.org/abs/1710.10467),
[Angular Loss](https://arxiv.org/pdf/2003.11982.pdf)
You can also help us implement more models.
<!-- start installation -->
## Installation
🐸TTS is tested on Ubuntu 18.04 with **python >= 3.9, < 3.12.**.
If you are only interested in [synthesizing speech](https://tts.readthedocs.io/en/latest/inference.html) with the released 🐸TTS models, installing from PyPI is the easiest option.
🐸TTS is tested on Ubuntu 24.04 with **python >= 3.9, < 3.13**, but should also
work on Mac and Windows.
If you are only interested in [synthesizing speech](https://coqui-tts.readthedocs.io/en/latest/inference.html) with the pretrained 🐸TTS models, installing from PyPI is the easiest option.
```bash
pip install TTS
pip install coqui-tts
```
If you plan to code or train models, clone 🐸TTS and install it locally.
```bash
git clone https://github.com/coqui-ai/TTS
pip install -e .[all,dev,notebooks] # Select the relevant extras
git clone https://github.com/idiap/coqui-ai-TTS
cd coqui-ai-TTS
pip install -e .
```
If you are on Ubuntu (Debian), you can also run following commands for installation.
### Optional dependencies
The following extras allow the installation of optional dependencies:
| Name | Description |
|------|-------------|
| `all` | All optional dependencies |
| `notebooks` | Dependencies only used in notebooks |
| `server` | Dependencies to run the TTS server |
| `bn` | Bangla G2P |
| `ja` | Japanese G2P |
| `ko` | Korean G2P |
| `zh` | Chinese G2P |
| `languages` | All language-specific dependencies |
You can install extras with one of the following commands:
```bash
$ make system-deps # intended to be used on Ubuntu (Debian). Let us know if you have a different OS.
$ make install
pip install coqui-tts[server,ja]
pip install -e .[server,ja]
```
If you are on Windows, 👑@GuyPaddock wrote installation instructions [here](https://stackoverflow.com/questions/66726331/how-can-i-run-mozilla-tts-coqui-tts-training-with-cuda-on-a-windows-system).
### Platforms
If you are on Ubuntu (Debian), you can also run the following commands for installation.
```bash
make system-deps
make install
```
<!-- end installation -->
## Docker Image
You can also try TTS without install with the docker image.
Simply run the following command and you will be able to run TTS without installing it.
You can also try out Coqui TTS without installation with the docker image.
Simply run the following command and you will be able to run TTS:
```bash
docker run --rm -it -p 5002:5002 --entrypoint /bin/bash ghcr.io/coqui-ai/tts-cpu
@ -180,14 +176,15 @@ python3 TTS/server/server.py --model_name tts_models/en/vctk/vits # To start a s
```
You can then enjoy the TTS server [here](http://[::1]:5002/)
More details about the docker images (like GPU support) can be found [here](https://tts.readthedocs.io/en/latest/docker_images.html)
More details about the docker images (like GPU support) can be found
[here](https://coqui-tts.readthedocs.io/en/latest/docker_images.html)
## Synthesizing speech by 🐸TTS
<!-- start inference -->
### 🐍 Python API
#### Running a multi-speaker and multi-lingual model
#### Multi-speaker and multi-lingual model
```python
import torch
@ -199,44 +196,63 @@ device = "cuda" if torch.cuda.is_available() else "cpu"
# List available 🐸TTS models
print(TTS().list_models())
# Init TTS
# Initialize TTS
tts = TTS("tts_models/multilingual/multi-dataset/xtts_v2").to(device)
# List speakers
print(tts.speakers)
# Run TTS
# ❗ Since this model is multi-lingual voice cloning model, we must set the target speaker_wav and language
# Text to speech list of amplitude values as output
wav = tts.tts(text="Hello world!", speaker_wav="my/cloning/audio.wav", language="en")
# Text to speech to a file
tts.tts_to_file(text="Hello world!", speaker_wav="my/cloning/audio.wav", language="en", file_path="output.wav")
# ❗ XTTS supports both, but many models allow only one of the `speaker` and
# `speaker_wav` arguments
# TTS with list of amplitude values as output, clone the voice from `speaker_wav`
wav = tts.tts(
text="Hello world!",
speaker_wav="my/cloning/audio.wav",
language="en"
)
# TTS to a file, use a preset speaker
tts.tts_to_file(
text="Hello world!",
speaker="Craig Gutsy",
language="en",
file_path="output.wav"
)
```
#### Running a single speaker model
#### Single speaker model
```python
# Init TTS with the target model name
tts = TTS(model_name="tts_models/de/thorsten/tacotron2-DDC", progress_bar=False).to(device)
# Initialize TTS with the target model name
tts = TTS("tts_models/de/thorsten/tacotron2-DDC").to(device)
# Run TTS
tts.tts_to_file(text="Ich bin eine Testnachricht.", file_path=OUTPUT_PATH)
# Example voice cloning with YourTTS in English, French and Portuguese
tts = TTS(model_name="tts_models/multilingual/multi-dataset/your_tts", progress_bar=False).to(device)
tts.tts_to_file("This is voice cloning.", speaker_wav="my/cloning/audio.wav", language="en", file_path="output.wav")
tts.tts_to_file("C'est le clonage de la voix.", speaker_wav="my/cloning/audio.wav", language="fr-fr", file_path="output.wav")
tts.tts_to_file("Isso é clonagem de voz.", speaker_wav="my/cloning/audio.wav", language="pt-br", file_path="output.wav")
```
#### Example voice conversion
#### Voice conversion (VC)
Converting the voice in `source_wav` to the voice of `target_wav`
```python
tts = TTS(model_name="voice_conversion_models/multilingual/vctk/freevc24", progress_bar=False).to("cuda")
tts.voice_conversion_to_file(source_wav="my/source.wav", target_wav="my/target.wav", file_path="output.wav")
tts = TTS("voice_conversion_models/multilingual/vctk/freevc24").to("cuda")
tts.voice_conversion_to_file(
source_wav="my/source.wav",
target_wav="my/target.wav",
file_path="output.wav"
)
```
#### Example voice cloning together with the voice conversion model.
This way, you can clone voices by using any model in 🐸TTS.
Other available voice conversion models:
- `voice_conversion_models/multilingual/multi-dataset/openvoice_v1`
- `voice_conversion_models/multilingual/multi-dataset/openvoice_v2`
#### Voice cloning by combining single speaker TTS model with the default VC model
This way, you can clone voices by using any model in 🐸TTS. The FreeVC model is
used for voice conversion after synthesizing speech.
```python
@ -248,160 +264,140 @@ tts.tts_with_vc_to_file(
)
```
#### Example text to speech using **Fairseq models in ~1100 languages** 🤯.
#### TTS using Fairseq models in ~1100 languages 🤯
For Fairseq models, use the following name format: `tts_models/<lang-iso_code>/fairseq/vits`.
You can find the language ISO codes [here](https://dl.fbaipublicfiles.com/mms/tts/all-tts-languages.html)
and learn about the Fairseq models [here](https://github.com/facebookresearch/fairseq/tree/main/examples/mms).
```python
# TTS with on the fly voice conversion
# TTS with fairseq models
api = TTS("tts_models/deu/fairseq/vits")
api.tts_with_vc_to_file(
api.tts_to_file(
"Wie sage ich auf Italienisch, dass ich dich liebe?",
speaker_wav="target/speaker.wav",
file_path="output.wav"
)
```
### Command-line `tts`
### Command-line interface `tts`
<!-- begin-tts-readme -->
Synthesize speech on command line.
Synthesize speech on the command line.
You can either use your trained model or choose a model from the provided list.
If you don't specify any models, then it uses LJSpeech based English model.
#### Single Speaker Models
- List provided models:
```
$ tts --list_models
```sh
tts --list_models
```
- Get model info (for both tts_models and vocoder_models):
- Query by type/name:
The model_info_by_name uses the name as it from the --list_models.
```
$ tts --model_info_by_name "<model_type>/<language>/<dataset>/<model_name>"
- Get model information. Use the names obtained from `--list_models`.
```sh
tts --model_info_by_name "<model_type>/<language>/<dataset>/<model_name>"
```
For example:
```
$ tts --model_info_by_name tts_models/tr/common-voice/glow-tts
$ tts --model_info_by_name vocoder_models/en/ljspeech/hifigan_v2
```
- Query by type/idx:
The model_query_idx uses the corresponding idx from --list_models.
```
$ tts --model_info_by_idx "<model_type>/<model_query_idx>"
```sh
tts --model_info_by_name tts_models/tr/common-voice/glow-tts
tts --model_info_by_name vocoder_models/en/ljspeech/hifigan_v2
```
For example:
#### Single speaker models
```
$ tts --model_info_by_idx tts_models/3
```
- Run TTS with the default model (`tts_models/en/ljspeech/tacotron2-DDC`):
- Query info for model info by full name:
```
$ tts --model_info_by_name "<model_type>/<language>/<dataset>/<model_name>"
```
- Run TTS with default models:
```
$ tts --text "Text for TTS" --out_path output/path/speech.wav
```sh
tts --text "Text for TTS" --out_path output/path/speech.wav
```
- Run TTS and pipe out the generated TTS wav file data:
```
$ tts --text "Text for TTS" --pipe_out --out_path output/path/speech.wav | aplay
```sh
tts --text "Text for TTS" --pipe_out --out_path output/path/speech.wav | aplay
```
- Run a TTS model with its default vocoder model:
```
$ tts --text "Text for TTS" --model_name "<model_type>/<language>/<dataset>/<model_name>" --out_path output/path/speech.wav
```sh
tts --text "Text for TTS" \
--model_name "<model_type>/<language>/<dataset>/<model_name>" \
--out_path output/path/speech.wav
```
For example:
```
$ tts --text "Text for TTS" --model_name "tts_models/en/ljspeech/glow-tts" --out_path output/path/speech.wav
```sh
tts --text "Text for TTS" \
--model_name "tts_models/en/ljspeech/glow-tts" \
--out_path output/path/speech.wav
```
- Run with specific TTS and vocoder models from the list:
- Run with specific TTS and vocoder models from the list. Note that not every vocoder is compatible with every TTS model.
```
$ tts --text "Text for TTS" --model_name "<model_type>/<language>/<dataset>/<model_name>" --vocoder_name "<model_type>/<language>/<dataset>/<model_name>" --out_path output/path/speech.wav
```sh
tts --text "Text for TTS" \
--model_name "<model_type>/<language>/<dataset>/<model_name>" \
--vocoder_name "<model_type>/<language>/<dataset>/<model_name>" \
--out_path output/path/speech.wav
```
For example:
```
$ tts --text "Text for TTS" --model_name "tts_models/en/ljspeech/glow-tts" --vocoder_name "vocoder_models/en/ljspeech/univnet" --out_path output/path/speech.wav
```sh
tts --text "Text for TTS" \
--model_name "tts_models/en/ljspeech/glow-tts" \
--vocoder_name "vocoder_models/en/ljspeech/univnet" \
--out_path output/path/speech.wav
```
- Run your own TTS model (Using Griffin-Lim Vocoder):
- Run your own TTS model (using Griffin-Lim Vocoder):
```
$ tts --text "Text for TTS" --model_path path/to/model.pth --config_path path/to/config.json --out_path output/path/speech.wav
```sh
tts --text "Text for TTS" \
--model_path path/to/model.pth \
--config_path path/to/config.json \
--out_path output/path/speech.wav
```
- Run your own TTS and Vocoder models:
```
$ tts --text "Text for TTS" --model_path path/to/model.pth --config_path path/to/config.json --out_path output/path/speech.wav
--vocoder_path path/to/vocoder.pth --vocoder_config_path path/to/vocoder_config.json
```sh
tts --text "Text for TTS" \
--model_path path/to/model.pth \
--config_path path/to/config.json \
--out_path output/path/speech.wav \
--vocoder_path path/to/vocoder.pth \
--vocoder_config_path path/to/vocoder_config.json
```
#### Multi-speaker Models
#### Multi-speaker models
- List the available speakers and choose a <speaker_id> among them:
- List the available speakers and choose a `<speaker_id>` among them:
```
$ tts --model_name "<language>/<dataset>/<model_name>" --list_speaker_idxs
```sh
tts --model_name "<language>/<dataset>/<model_name>" --list_speaker_idxs
```
- Run the multi-speaker TTS model with the target speaker ID:
```
$ tts --text "Text for TTS." --out_path output/path/speech.wav --model_name "<language>/<dataset>/<model_name>" --speaker_idx <speaker_id>
```sh
tts --text "Text for TTS." --out_path output/path/speech.wav \
--model_name "<language>/<dataset>/<model_name>" --speaker_idx <speaker_id>
```
- Run your own multi-speaker TTS model:
```
$ tts --text "Text for TTS" --out_path output/path/speech.wav --model_path path/to/model.pth --config_path path/to/config.json --speakers_file_path path/to/speaker.json --speaker_idx <speaker_id>
```sh
tts --text "Text for TTS" --out_path output/path/speech.wav \
--model_path path/to/model.pth --config_path path/to/config.json \
--speakers_file_path path/to/speaker.json --speaker_idx <speaker_id>
```
### Voice Conversion Models
#### Voice conversion models
```
$ tts --out_path output/path/speech.wav --model_name "<language>/<dataset>/<model_name>" --source_wav <path/to/speaker/wav> --target_wav <path/to/reference/wav>
```sh
tts --out_path output/path/speech.wav --model_name "<language>/<dataset>/<model_name>" \
--source_wav <path/to/speaker/wav> --target_wav <path/to/reference/wav>
```
<!-- end-tts-readme -->
## Directory Structure
```
|- notebooks/ (Jupyter Notebooks for model evaluation, parameter selection and data analysis.)
|- utils/ (common utilities.)
|- TTS
|- bin/ (folder for all the executables.)
|- train*.py (train your target model.)
|- ...
|- tts/ (text to speech models)
|- layers/ (model layer definitions)
|- models/ (model definitions)
|- utils/ (model specific utilities.)
|- speaker_encoder/ (Speaker Encoder models.)
|- (same)
|- vocoder/ (Vocoder models.)
|- (same)
```

237
TTS/.models.json
View File

@ -5,11 +5,11 @@
"xtts_v2": {
"description": "XTTS-v2.0.3 by Coqui with 17 languages.",
"hf_url": [
"https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v2/main/model.pth",
"https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v2/main/config.json",
"https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v2/main/vocab.json",
"https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v2/main/hash.md5",
"https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v2/main/speakers_xtts.pth"
"https://huggingface.co/coqui/XTTS-v2/resolve/main/model.pth",
"https://huggingface.co/coqui/XTTS-v2/resolve/main/config.json",
"https://huggingface.co/coqui/XTTS-v2/resolve/main/vocab.json",
"https://huggingface.co/coqui/XTTS-v2/resolve/main/hash.md5",
"https://huggingface.co/coqui/XTTS-v2/resolve/main/speakers_xtts.pth"
],
"model_hash": "10f92b55c512af7a8d39d650547a15a7",
"default_vocoder": null,
@ -21,10 +21,10 @@
"xtts_v1.1": {
"description": "XTTS-v1.1 by Coqui with 14 languages, cross-language voice cloning and reference leak fixed.",
"hf_url": [
"https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v1/v1.1.2/model.pth",
"https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v1/v1.1.2/config.json",
"https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v1/v1.1.2/vocab.json",
"https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v1/v1.1.2/hash.md5"
"https://huggingface.co/coqui/XTTS-v1/resolve/v1.1.2/model.pth",
"https://huggingface.co/coqui/XTTS-v1/resolve/v1.1.2/config.json",
"https://huggingface.co/coqui/XTTS-v1/resolve/v1.1.2/vocab.json",
"https://huggingface.co/coqui/XTTS-v1/resolve/v1.1.2/hash.md5"
],
"model_hash": "7c62beaf58d39b729de287330dc254e7b515677416839b649a50e7cf74c3df59",
"default_vocoder": null,
@ -35,7 +35,7 @@
},
"your_tts": {
"description": "Your TTS model accompanying the paper https://arxiv.org/abs/2112.02418",
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.10.1_models/tts_models--multilingual--multi-dataset--your_tts.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.10.1_models/tts_models--multilingual--multi-dataset--your_tts.zip",
"default_vocoder": null,
"commit": "e9a1953e",
"license": "CC BY-NC-ND 4.0",
@ -44,12 +44,11 @@
"bark": {
"description": "🐶 Bark TTS model released by suno-ai. You can find the original implementation in https://github.com/suno-ai/bark.",
"hf_url": [
"https://coqui.gateway.scarf.sh/hf/bark/coarse_2.pt",
"https://coqui.gateway.scarf.sh/hf/bark/fine_2.pt",
"https://coqui.gateway.scarf.sh/hf/text_2.pt",
"https://coqui.gateway.scarf.sh/hf/bark/config.json",
"https://coqui.gateway.scarf.sh/hf/bark/hubert.pt",
"https://coqui.gateway.scarf.sh/hf/bark/tokenizer.pth"
"https://huggingface.co/erogol/bark/resolve/main/coarse_2.pt",
"https://huggingface.co/erogol/bark/resolve/main/fine_2.pt",
"https://huggingface.co/erogol/bark/resolve/main/text_2.pt",
"https://huggingface.co/erogol/bark/resolve/main/config.json",
"https://huggingface.co/erogol/bark/resolve/main/tokenizer.pth"
],
"default_vocoder": null,
"commit": "e9a1953e",
@ -61,7 +60,7 @@
"bg": {
"cv": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--bg--cv--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/tts_models--bg--cv--vits.zip",
"default_vocoder": null,
"commit": null,
"author": "@NeonGeckoCom",
@ -72,7 +71,7 @@
"cs": {
"cv": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--cs--cv--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/tts_models--cs--cv--vits.zip",
"default_vocoder": null,
"commit": null,
"author": "@NeonGeckoCom",
@ -83,7 +82,7 @@
"da": {
"cv": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--da--cv--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/tts_models--da--cv--vits.zip",
"default_vocoder": null,
"commit": null,
"author": "@NeonGeckoCom",
@ -94,7 +93,7 @@
"et": {
"cv": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--et--cv--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/tts_models--et--cv--vits.zip",
"default_vocoder": null,
"commit": null,
"author": "@NeonGeckoCom",
@ -105,7 +104,7 @@
"ga": {
"cv": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--ga--cv--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/tts_models--ga--cv--vits.zip",
"default_vocoder": null,
"commit": null,
"author": "@NeonGeckoCom",
@ -117,7 +116,7 @@
"ek1": {
"tacotron2": {
"description": "EK1 en-rp tacotron2 by NMStoker",
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--en--ek1--tacotron2.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/tts_models--en--ek1--tacotron2.zip",
"default_vocoder": "vocoder_models/en/ek1/wavegrad",
"commit": "c802255",
"license": "apache 2.0"
@ -126,7 +125,7 @@
"ljspeech": {
"tacotron2-DDC": {
"description": "Tacotron2 with Double Decoder Consistency.",
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--en--ljspeech--tacotron2-DDC.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/tts_models--en--ljspeech--tacotron2-DDC.zip",
"default_vocoder": "vocoder_models/en/ljspeech/hifigan_v2",
"commit": "bae2ad0f",
"author": "Eren Gölge @erogol",
@ -135,7 +134,7 @@
},
"tacotron2-DDC_ph": {
"description": "Tacotron2 with Double Decoder Consistency with phonemes.",
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--en--ljspeech--tacotron2-DDC_ph.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/tts_models--en--ljspeech--tacotron2-DDC_ph.zip",
"default_vocoder": "vocoder_models/en/ljspeech/univnet",
"commit": "3900448",
"author": "Eren Gölge @erogol",
@ -144,7 +143,7 @@
},
"glow-tts": {
"description": "",
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--en--ljspeech--glow-tts.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/tts_models--en--ljspeech--glow-tts.zip",
"stats_file": null,
"default_vocoder": "vocoder_models/en/ljspeech/multiband-melgan",
"commit": "",
@ -154,7 +153,7 @@
},
"speedy-speech": {
"description": "Speedy Speech model trained on LJSpeech dataset using the Alignment Network for learning the durations.",
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--en--ljspeech--speedy-speech.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/tts_models--en--ljspeech--speedy-speech.zip",
"stats_file": null,
"default_vocoder": "vocoder_models/en/ljspeech/hifigan_v2",
"commit": "4581e3d",
@ -164,7 +163,7 @@
},
"tacotron2-DCA": {
"description": "",
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--en--ljspeech--tacotron2-DCA.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/tts_models--en--ljspeech--tacotron2-DCA.zip",
"default_vocoder": "vocoder_models/en/ljspeech/multiband-melgan",
"commit": "",
"author": "Eren Gölge @erogol",
@ -173,7 +172,7 @@
},
"vits": {
"description": "VITS is an End2End TTS model trained on LJSpeech dataset with phonemes.",
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--en--ljspeech--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/tts_models--en--ljspeech--vits.zip",
"default_vocoder": null,
"commit": "3900448",
"author": "Eren Gölge @erogol",
@ -181,7 +180,7 @@
"contact": "egolge@coqui.com"
},
"vits--neon": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--en--ljspeech--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/tts_models--en--ljspeech--vits.zip",
"default_vocoder": null,
"author": "@NeonGeckoCom",
"license": "bsd-3-clause",
@ -190,7 +189,7 @@
},
"fast_pitch": {
"description": "FastPitch model trained on LJSpeech using the Aligner Network",
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--en--ljspeech--fast_pitch.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/tts_models--en--ljspeech--fast_pitch.zip",
"default_vocoder": "vocoder_models/en/ljspeech/hifigan_v2",
"commit": "b27b3ba",
"author": "Eren Gölge @erogol",
@ -199,7 +198,7 @@
},
"overflow": {
"description": "Overflow model trained on LJSpeech",
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.10.0_models/tts_models--en--ljspeech--overflow.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.10.0_models/tts_models--en--ljspeech--overflow.zip",
"default_vocoder": "vocoder_models/en/ljspeech/hifigan_v2",
"commit": "3b1a28f",
"author": "Eren Gölge @erogol",
@ -208,7 +207,7 @@
},
"neural_hmm": {
"description": "Neural HMM model trained on LJSpeech",
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.11.0_models/tts_models--en--ljspeech--neural_hmm.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.11.0_models/tts_models--en--ljspeech--neural_hmm.zip",
"default_vocoder": "vocoder_models/en/ljspeech/hifigan_v2",
"commit": "3b1a28f",
"author": "Shivam Metha @shivammehta25",
@ -219,7 +218,7 @@
"vctk": {
"vits": {
"description": "VITS End2End TTS model trained on VCTK dataset with 109 different speakers with EN accent.",
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--en--vctk--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/tts_models--en--vctk--vits.zip",
"default_vocoder": null,
"commit": "3900448",
"author": "Eren @erogol",
@ -228,7 +227,7 @@
},
"fast_pitch": {
"description": "FastPitch model trained on VCTK dataseset.",
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--en--vctk--fast_pitch.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/tts_models--en--vctk--fast_pitch.zip",
"default_vocoder": null,
"commit": "bdab788d",
"author": "Eren @erogol",
@ -239,7 +238,7 @@
"sam": {
"tacotron-DDC": {
"description": "Tacotron2 with Double Decoder Consistency trained with Aceenture's Sam dataset.",
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--en--sam--tacotron-DDC.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/tts_models--en--sam--tacotron-DDC.zip",
"default_vocoder": "vocoder_models/en/sam/hifigan_v2",
"commit": "bae2ad0f",
"author": "Eren Gölge @erogol",
@ -250,7 +249,7 @@
"blizzard2013": {
"capacitron-t2-c50": {
"description": "Capacitron additions to Tacotron 2 with Capacity at 50 as in https://arxiv.org/pdf/1906.03402.pdf",
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.7.0_models/tts_models--en--blizzard2013--capacitron-t2-c50.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.7.0_models/tts_models--en--blizzard2013--capacitron-t2-c50.zip",
"commit": "d6284e7",
"default_vocoder": "vocoder_models/en/blizzard2013/hifigan_v2",
"author": "Adam Froghyar @a-froghyar",
@ -259,7 +258,7 @@
},
"capacitron-t2-c150_v2": {
"description": "Capacitron additions to Tacotron 2 with Capacity at 150 as in https://arxiv.org/pdf/1906.03402.pdf",
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.7.1_models/tts_models--en--blizzard2013--capacitron-t2-c150_v2.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.7.1_models/tts_models--en--blizzard2013--capacitron-t2-c150_v2.zip",
"commit": "a67039d",
"default_vocoder": "vocoder_models/en/blizzard2013/hifigan_v2",
"author": "Adam Froghyar @a-froghyar",
@ -271,15 +270,15 @@
"tortoise-v2": {
"description": "Tortoise tts model https://github.com/neonbjb/tortoise-tts",
"github_rls_url": [
"https://coqui.gateway.scarf.sh/v0.14.1_models/autoregressive.pth",
"https://coqui.gateway.scarf.sh/v0.14.1_models/clvp2.pth",
"https://coqui.gateway.scarf.sh/v0.14.1_models/cvvp.pth",
"https://coqui.gateway.scarf.sh/v0.14.1_models/diffusion_decoder.pth",
"https://coqui.gateway.scarf.sh/v0.14.1_models/rlg_auto.pth",
"https://coqui.gateway.scarf.sh/v0.14.1_models/rlg_diffuser.pth",
"https://coqui.gateway.scarf.sh/v0.14.1_models/vocoder.pth",
"https://coqui.gateway.scarf.sh/v0.14.1_models/mel_norms.pth",
"https://coqui.gateway.scarf.sh/v0.14.1_models/config.json"
"https://github.com/coqui-ai/TTS/releases/download/v0.14.1_models/autoregressive.pth",
"https://github.com/coqui-ai/TTS/releases/download/v0.14.1_models/clvp2.pth",
"https://github.com/coqui-ai/TTS/releases/download/v0.14.1_models/cvvp.pth",
"https://github.com/coqui-ai/TTS/releases/download/v0.14.1_models/diffusion_decoder.pth",
"https://github.com/coqui-ai/TTS/releases/download/v0.14.1_models/rlg_auto.pth",
"https://github.com/coqui-ai/TTS/releases/download/v0.14.1_models/rlg_diffuser.pth",
"https://github.com/coqui-ai/TTS/releases/download/v0.14.1_models/vocoder.pth",
"https://github.com/coqui-ai/TTS/releases/download/v0.14.1_models/mel_norms.pth",
"https://github.com/coqui-ai/TTS/releases/download/v0.14.1_models/config.json"
],
"commit": "c1875f6",
"default_vocoder": null,
@ -290,7 +289,7 @@
"jenny": {
"jenny": {
"description": "VITS model trained with Jenny(Dioco) dataset. Named as Jenny as demanded by the license. Original URL for the model https://www.kaggle.com/datasets/noml4u/tts-models--en--jenny-dioco--vits",
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.14.0_models/tts_models--en--jenny--jenny.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.14.0_models/tts_models--en--jenny--jenny.zip",
"default_vocoder": null,
"commit": "ba40a1c",
"license": "custom - see https://github.com/dioco-group/jenny-tts-dataset#important",
@ -301,7 +300,7 @@
"es": {
"mai": {
"tacotron2-DDC": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--es--mai--tacotron2-DDC.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/tts_models--es--mai--tacotron2-DDC.zip",
"default_vocoder": "vocoder_models/universal/libri-tts/fullband-melgan",
"commit": "",
"author": "Eren Gölge @erogol",
@ -311,7 +310,7 @@
},
"css10": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--es--css10--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/tts_models--es--css10--vits.zip",
"default_vocoder": null,
"commit": null,
"author": "@NeonGeckoCom",
@ -322,7 +321,7 @@
"fr": {
"mai": {
"tacotron2-DDC": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--fr--mai--tacotron2-DDC.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/tts_models--fr--mai--tacotron2-DDC.zip",
"default_vocoder": "vocoder_models/universal/libri-tts/fullband-melgan",
"commit": null,
"author": "Eren Gölge @erogol",
@ -332,7 +331,7 @@
},
"css10": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--fr--css10--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/tts_models--fr--css10--vits.zip",
"default_vocoder": null,
"commit": null,
"author": "@NeonGeckoCom",
@ -343,7 +342,7 @@
"uk": {
"mai": {
"glow-tts": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--uk--mai--glow-tts.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/tts_models--uk--mai--glow-tts.zip",
"author": "@robinhad",
"commit": "bdab788d",
"license": "MIT",
@ -351,7 +350,7 @@
"default_vocoder": "vocoder_models/uk/mai/multiband-melgan"
},
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--uk--mai--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/tts_models--uk--mai--vits.zip",
"default_vocoder": null,
"commit": null,
"author": "@NeonGeckoCom",
@ -362,7 +361,7 @@
"zh-CN": {
"baker": {
"tacotron2-DDC-GST": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--zh-CN--baker--tacotron2-DDC-GST.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/tts_models--zh-CN--baker--tacotron2-DDC-GST.zip",
"commit": "unknown",
"author": "@kirianguiller",
"license": "apache 2.0",
@ -373,7 +372,7 @@
"nl": {
"mai": {
"tacotron2-DDC": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--nl--mai--tacotron2-DDC.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/tts_models--nl--mai--tacotron2-DDC.zip",
"author": "@r-dh",
"license": "apache 2.0",
"default_vocoder": "vocoder_models/nl/mai/parallel-wavegan",
@ -383,7 +382,7 @@
},
"css10": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--nl--css10--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/tts_models--nl--css10--vits.zip",
"default_vocoder": null,
"commit": null,
"author": "@NeonGeckoCom",
@ -394,21 +393,21 @@
"de": {
"thorsten": {
"tacotron2-DCA": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--de--thorsten--tacotron2-DCA.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/tts_models--de--thorsten--tacotron2-DCA.zip",
"default_vocoder": "vocoder_models/de/thorsten/fullband-melgan",
"author": "@thorstenMueller",
"license": "apache 2.0",
"commit": "unknown"
},
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.7.0_models/tts_models--de--thorsten--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.7.0_models/tts_models--de--thorsten--vits.zip",
"default_vocoder": null,
"author": "@thorstenMueller",
"license": "apache 2.0",
"commit": "unknown"
},
"tacotron2-DDC": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--de--thorsten--tacotron2-DDC.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/tts_models--de--thorsten--tacotron2-DDC.zip",
"default_vocoder": "vocoder_models/de/thorsten/hifigan_v1",
"description": "Thorsten-Dec2021-22k-DDC",
"author": "@thorstenMueller",
@ -418,7 +417,7 @@
},
"css10": {
"vits-neon": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--de--css10--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/tts_models--de--css10--vits.zip",
"default_vocoder": null,
"author": "@NeonGeckoCom",
"license": "bsd-3-clause",
@ -429,7 +428,7 @@
"ja": {
"kokoro": {
"tacotron2-DDC": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--ja--kokoro--tacotron2-DDC.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/tts_models--ja--kokoro--tacotron2-DDC.zip",
"default_vocoder": "vocoder_models/ja/kokoro/hifigan_v1",
"description": "Tacotron2 with Double Decoder Consistency trained with Kokoro Speech Dataset.",
"author": "@kaiidams",
@ -441,7 +440,7 @@
"tr": {
"common-voice": {
"glow-tts": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--tr--common-voice--glow-tts.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/tts_models--tr--common-voice--glow-tts.zip",
"default_vocoder": "vocoder_models/tr/common-voice/hifigan",
"license": "MIT",
"description": "Turkish GlowTTS model using an unknown speaker from the Common-Voice dataset.",
@ -453,7 +452,7 @@
"it": {
"mai_female": {
"glow-tts": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--it--mai_female--glow-tts.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/tts_models--it--mai_female--glow-tts.zip",
"default_vocoder": null,
"description": "GlowTTS model as explained on https://github.com/coqui-ai/TTS/issues/1148.",
"author": "@nicolalandro",
@ -461,7 +460,7 @@
"commit": null
},
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--it--mai_female--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/tts_models--it--mai_female--vits.zip",
"default_vocoder": null,
"description": "GlowTTS model as explained on https://github.com/coqui-ai/TTS/issues/1148.",
"author": "@nicolalandro",
@ -471,7 +470,7 @@
},
"mai_male": {
"glow-tts": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--it--mai_male--glow-tts.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/tts_models--it--mai_male--glow-tts.zip",
"default_vocoder": null,
"description": "GlowTTS model as explained on https://github.com/coqui-ai/TTS/issues/1148.",
"author": "@nicolalandro",
@ -479,7 +478,7 @@
"commit": null
},
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/tts_models--it--mai_male--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/tts_models--it--mai_male--vits.zip",
"default_vocoder": null,
"description": "GlowTTS model as explained on https://github.com/coqui-ai/TTS/issues/1148.",
"author": "@nicolalandro",
@ -491,7 +490,7 @@
"ewe": {
"openbible": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.2_models/tts_models--ewe--openbible--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.2_models/tts_models--ewe--openbible--vits.zip",
"default_vocoder": null,
"license": "CC-BY-SA 4.0",
"description": "Original work (audio and text) by Biblica available for free at www.biblica.com and open.bible.",
@ -503,7 +502,7 @@
"hau": {
"openbible": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.2_models/tts_models--hau--openbible--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.2_models/tts_models--hau--openbible--vits.zip",
"default_vocoder": null,
"license": "CC-BY-SA 4.0",
"description": "Original work (audio and text) by Biblica available for free at www.biblica.com and open.bible.",
@ -515,7 +514,7 @@
"lin": {
"openbible": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.2_models/tts_models--lin--openbible--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.2_models/tts_models--lin--openbible--vits.zip",
"default_vocoder": null,
"license": "CC-BY-SA 4.0",
"description": "Original work (audio and text) by Biblica available for free at www.biblica.com and open.bible.",
@ -527,7 +526,7 @@
"tw_akuapem": {
"openbible": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.2_models/tts_models--tw_akuapem--openbible--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.2_models/tts_models--tw_akuapem--openbible--vits.zip",
"default_vocoder": null,
"license": "CC-BY-SA 4.0",
"description": "Original work (audio and text) by Biblica available for free at www.biblica.com and open.bible.",
@ -539,7 +538,7 @@
"tw_asante": {
"openbible": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.2_models/tts_models--tw_asante--openbible--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.2_models/tts_models--tw_asante--openbible--vits.zip",
"default_vocoder": null,
"license": "CC-BY-SA 4.0",
"description": "Original work (audio and text) by Biblica available for free at www.biblica.com and open.bible.",
@ -551,7 +550,7 @@
"yor": {
"openbible": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.2_models/tts_models--yor--openbible--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.2_models/tts_models--yor--openbible--vits.zip",
"default_vocoder": null,
"license": "CC-BY-SA 4.0",
"description": "Original work (audio and text) by Biblica available for free at www.biblica.com and open.bible.",
@ -563,7 +562,7 @@
"hu": {
"css10": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--hu--css10--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/tts_models--hu--css10--vits.zip",
"default_vocoder": null,
"commit": null,
"author": "@NeonGeckoCom",
@ -574,7 +573,7 @@
"el": {
"cv": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--el--cv--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/tts_models--el--cv--vits.zip",
"default_vocoder": null,
"commit": null,
"author": "@NeonGeckoCom",
@ -585,7 +584,7 @@
"fi": {
"css10": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--fi--css10--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/tts_models--fi--css10--vits.zip",
"default_vocoder": null,
"commit": null,
"author": "@NeonGeckoCom",
@ -596,7 +595,7 @@
"hr": {
"cv": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--hr--cv--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/tts_models--hr--cv--vits.zip",
"default_vocoder": null,
"commit": null,
"author": "@NeonGeckoCom",
@ -607,7 +606,7 @@
"lt": {
"cv": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--lt--cv--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/tts_models--lt--cv--vits.zip",
"default_vocoder": null,
"commit": null,
"author": "@NeonGeckoCom",
@ -618,7 +617,7 @@
"lv": {
"cv": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--lv--cv--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/tts_models--lv--cv--vits.zip",
"default_vocoder": null,
"commit": null,
"author": "@NeonGeckoCom",
@ -629,7 +628,7 @@
"mt": {
"cv": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--mt--cv--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/tts_models--mt--cv--vits.zip",
"default_vocoder": null,
"commit": null,
"author": "@NeonGeckoCom",
@ -640,7 +639,7 @@
"pl": {
"mai_female": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--pl--mai_female--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/tts_models--pl--mai_female--vits.zip",
"default_vocoder": null,
"commit": null,
"author": "@NeonGeckoCom",
@ -651,7 +650,7 @@
"pt": {
"cv": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--pt--cv--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/tts_models--pt--cv--vits.zip",
"default_vocoder": null,
"commit": null,
"author": "@NeonGeckoCom",
@ -662,7 +661,7 @@
"ro": {
"cv": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--ro--cv--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/tts_models--ro--cv--vits.zip",
"default_vocoder": null,
"commit": null,
"author": "@NeonGeckoCom",
@ -673,7 +672,7 @@
"sk": {
"cv": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--sk--cv--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/tts_models--sk--cv--vits.zip",
"default_vocoder": null,
"commit": null,
"author": "@NeonGeckoCom",
@ -684,7 +683,7 @@
"sl": {
"cv": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--sl--cv--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/tts_models--sl--cv--vits.zip",
"default_vocoder": null,
"commit": null,
"author": "@NeonGeckoCom",
@ -695,7 +694,7 @@
"sv": {
"cv": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/tts_models--sv--cv--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/tts_models--sv--cv--vits.zip",
"default_vocoder": null,
"commit": null,
"author": "@NeonGeckoCom",
@ -706,7 +705,7 @@
"ca": {
"custom": {
"vits": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.10.1_models/tts_models--ca--custom--vits.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.10.1_models/tts_models--ca--custom--vits.zip",
"default_vocoder": null,
"commit": null,
"description": " It is trained from zero with 101460 utterances consisting of 257 speakers, approx 138 hours of speech. We used three datasets;\nFestcat and Google Catalan TTS (both TTS datasets) and also a part of Common Voice 8. It is trained with TTS v0.8.0.\nhttps://github.com/coqui-ai/TTS/discussions/930#discussioncomment-4466345",
@ -718,7 +717,7 @@
"fa": {
"custom": {
"glow-tts": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.10.1_models/tts_models--fa--custom--glow-tts.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.10.1_models/tts_models--fa--custom--glow-tts.zip",
"default_vocoder": null,
"commit": null,
"description": "persian-tts-female-glow_tts model for text to speech purposes. Single-speaker female voice Trained on persian-tts-dataset-famale. \nThis model has no compatible vocoder thus the output quality is not very good. \nDataset: https://www.kaggle.com/datasets/magnoliasis/persian-tts-dataset-famale.",
@ -730,7 +729,7 @@
"bn": {
"custom": {
"vits-male": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.13.3_models/tts_models--bn--custom--vits_male.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.13.3_models/tts_models--bn--custom--vits_male.zip",
"default_vocoder": null,
"commit": null,
"description": "Single speaker Bangla male model. For more information -> https://github.com/mobassir94/comprehensive-bangla-tts",
@ -738,7 +737,7 @@
"license": "Apache 2.0"
},
"vits-female": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.13.3_models/tts_models--bn--custom--vits_female.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.13.3_models/tts_models--bn--custom--vits_female.zip",
"default_vocoder": null,
"commit": null,
"description": "Single speaker Bangla female model. For more information -> https://github.com/mobassir94/comprehensive-bangla-tts",
@ -751,7 +750,7 @@
"common-voice": {
"glow-tts":{
"description": "Belarusian GlowTTS model created by @alex73 (Github).",
"github_rls_url":"https://coqui.gateway.scarf.sh/v0.16.6/tts_models--be--common-voice--glow-tts.zip",
"github_rls_url":"https://github.com/coqui-ai/TTS/releases/download/v0.16.6/tts_models--be--common-voice--glow-tts.zip",
"default_vocoder": "vocoder_models/be/common-voice/hifigan",
"commit": "c0aabb85",
"license": "CC-BY-SA 4.0",
@ -764,14 +763,14 @@
"universal": {
"libri-tts": {
"wavegrad": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--universal--libri-tts--wavegrad.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/vocoder_models--universal--libri-tts--wavegrad.zip",
"commit": "ea976b0",
"author": "Eren Gölge @erogol",
"license": "MPL",
"contact": "egolge@coqui.com"
},
"fullband-melgan": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--universal--libri-tts--fullband-melgan.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/vocoder_models--universal--libri-tts--fullband-melgan.zip",
"commit": "4132240",
"author": "Eren Gölge @erogol",
"license": "MPL",
@ -783,14 +782,14 @@
"ek1": {
"wavegrad": {
"description": "EK1 en-rp wavegrad by NMStoker",
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--en--ek1--wavegrad.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/vocoder_models--en--ek1--wavegrad.zip",
"commit": "c802255",
"license": "apache 2.0"
}
},
"ljspeech": {
"multiband-melgan": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--en--ljspeech--multiband-melgan.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/vocoder_models--en--ljspeech--multiband-melgan.zip",
"commit": "ea976b0",
"author": "Eren Gölge @erogol",
"license": "MPL",
@ -798,7 +797,7 @@
},
"hifigan_v2": {
"description": "HiFiGAN_v2 LJSpeech vocoder from https://arxiv.org/abs/2010.05646.",
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--en--ljspeech--hifigan_v2.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/vocoder_models--en--ljspeech--hifigan_v2.zip",
"commit": "bae2ad0f",
"author": "@erogol",
"license": "apache 2.0",
@ -806,7 +805,7 @@
},
"univnet": {
"description": "UnivNet model finetuned on TacotronDDC_ph spectrograms for better compatibility.",
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--en--ljspeech--univnet_v2.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/vocoder_models--en--ljspeech--univnet_v2.zip",
"commit": "4581e3d",
"author": "Eren @erogol",
"license": "apache 2.0",
@ -816,7 +815,7 @@
"blizzard2013": {
"hifigan_v2": {
"description": "HiFiGAN_v2 LJSpeech vocoder from https://arxiv.org/abs/2010.05646.",
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.7.0_models/vocoder_models--en--blizzard2013--hifigan_v2.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.7.0_models/vocoder_models--en--blizzard2013--hifigan_v2.zip",
"commit": "d6284e7",
"author": "Adam Froghyar @a-froghyar",
"license": "apache 2.0",
@ -826,7 +825,7 @@
"vctk": {
"hifigan_v2": {
"description": "Finetuned and intended to be used with tts_models/en/vctk/sc-glow-tts",
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--en--vctk--hifigan_v2.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/vocoder_models--en--vctk--hifigan_v2.zip",
"commit": "2f07160",
"author": "Edresson Casanova",
"license": "apache 2.0",
@ -836,7 +835,7 @@
"sam": {
"hifigan_v2": {
"description": "Finetuned and intended to be used with tts_models/en/sam/tacotron_DDC",
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--en--sam--hifigan_v2.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/vocoder_models--en--sam--hifigan_v2.zip",
"commit": "2f07160",
"author": "Eren Gölge @erogol",
"license": "apache 2.0",
@ -847,7 +846,7 @@
"nl": {
"mai": {
"parallel-wavegan": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--nl--mai--parallel-wavegan.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/vocoder_models--nl--mai--parallel-wavegan.zip",
"author": "@r-dh",
"license": "apache 2.0",
"commit": "unknown"
@ -857,19 +856,19 @@
"de": {
"thorsten": {
"wavegrad": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--de--thorsten--wavegrad.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/vocoder_models--de--thorsten--wavegrad.zip",
"author": "@thorstenMueller",
"license": "apache 2.0",
"commit": "unknown"
},
"fullband-melgan": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--de--thorsten--fullband-melgan.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/vocoder_models--de--thorsten--fullband-melgan.zip",
"author": "@thorstenMueller",
"license": "apache 2.0",
"commit": "unknown"
},
"hifigan_v1": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.8.0_models/vocoder_models--de--thorsten--hifigan_v1.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.8.0_models/vocoder_models--de--thorsten--hifigan_v1.zip",
"description": "HifiGAN vocoder model for Thorsten Neutral Dec2021 22k Samplerate Tacotron2 DDC model",
"author": "@thorstenMueller",
"license": "apache 2.0",
@ -880,7 +879,7 @@
"ja": {
"kokoro": {
"hifigan_v1": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--ja--kokoro--hifigan_v1.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/vocoder_models--ja--kokoro--hifigan_v1.zip",
"description": "HifiGAN model trained for kokoro dataset by @kaiidams",
"author": "@kaiidams",
"license": "apache 2.0",
@ -891,7 +890,7 @@
"uk": {
"mai": {
"multiband-melgan": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--uk--mai--multiband-melgan.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/vocoder_models--uk--mai--multiband-melgan.zip",
"author": "@robinhad",
"commit": "bdab788d",
"license": "MIT",
@ -902,7 +901,7 @@
"tr": {
"common-voice": {
"hifigan": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.6.1_models/vocoder_models--tr--common-voice--hifigan.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.6.1_models/vocoder_models--tr--common-voice--hifigan.zip",
"description": "HifiGAN model using an unknown speaker from the Common-Voice dataset.",
"author": "Fatih Akademi",
"license": "MIT",
@ -913,7 +912,7 @@
"be": {
"common-voice": {
"hifigan": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.16.6/vocoder_models--be--common-voice--hifigan.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.16.6/vocoder_models--be--common-voice--hifigan.zip",
"description": "Belarusian HiFiGAN model created by @alex73 (Github).",
"author": "@alex73",
"license": "CC-BY-SA 4.0",
@ -926,12 +925,34 @@
"multilingual": {
"vctk": {
"freevc24": {
"github_rls_url": "https://coqui.gateway.scarf.sh/v0.13.0_models/voice_conversion_models--multilingual--vctk--freevc24.zip",
"github_rls_url": "https://github.com/coqui-ai/TTS/releases/download/v0.13.0_models/voice_conversion_models--multilingual--vctk--freevc24.zip",
"description": "FreeVC model trained on VCTK dataset from https://github.com/OlaWod/FreeVC",
"author": "Jing-Yi Li @OlaWod",
"license": "MIT",
"commit": null
}
},
"multi-dataset": {
"openvoice_v1": {
"hf_url": [
"https://huggingface.co/myshell-ai/OpenVoice/resolve/main/checkpoints/converter/config.json",
"https://huggingface.co/myshell-ai/OpenVoice/resolve/main/checkpoints/converter/checkpoint.pth"
],
"description": "OpenVoice VC model from https://huggingface.co/myshell-ai/OpenVoiceV2",
"author": "MyShell.ai",
"license": "MIT",
"commit": null
},
"openvoice_v2": {
"hf_url": [
"https://huggingface.co/myshell-ai/OpenVoiceV2/resolve/main/converter/config.json",
"https://huggingface.co/myshell-ai/OpenVoiceV2/resolve/main/converter/checkpoint.pth"
],
"description": "OpenVoice VC model from https://huggingface.co/myshell-ai/OpenVoiceV2",
"author": "MyShell.ai",
"license": "MIT",
"commit": null
}
}
}
}

1
TTS/VERSION
View File

@ -1 +0,0 @@
0.22.0

35
TTS/__init__.py
View File

@ -1,6 +1,33 @@
import os
import importlib.metadata
with open(os.path.join(os.path.dirname(__file__), "VERSION"), "r", encoding="utf-8") as f:
version = f.read().strip()
from TTS.utils.generic_utils import is_pytorch_at_least_2_4
__version__ = version
__version__ = importlib.metadata.version("coqui-tts")
if is_pytorch_at_least_2_4():
import _codecs
from collections import defaultdict
import numpy as np
import torch
from TTS.config.shared_configs import BaseDatasetConfig
from TTS.tts.configs.xtts_config import XttsConfig
from TTS.tts.models.xtts import XttsArgs, XttsAudioConfig
from TTS.utils.radam import RAdam
torch.serialization.add_safe_globals([dict, defaultdict, RAdam])
# Bark
torch.serialization.add_safe_globals(
[
np.core.multiarray.scalar,
np.dtype,
np.dtypes.Float64DType,
_codecs.encode, # TODO: safe by default from Pytorch 2.5
]
)
# XTTS
torch.serialization.add_safe_globals([BaseDatasetConfig, XttsConfig, XttsAudioConfig, XttsArgs])

197
TTS/api.py
View File

@ -1,15 +1,18 @@
"""Coqui TTS Python API."""
import logging
import tempfile
import warnings
from pathlib import Path
from typing import Union
from typing import Optional
import numpy as np
from torch import nn
from TTS.utils.audio.numpy_transforms import save_wav
from TTS.config import load_config
from TTS.utils.manage import ModelManager
from TTS.utils.synthesizer import Synthesizer
from TTS.config import load_config
logger = logging.getLogger(__name__)
class TTS(nn.Module):
@ -18,13 +21,19 @@ class TTS(nn.Module):
def __init__(
self,
model_name: str = "",
model_path: str = None,
config_path: str = None,
vocoder_path: str = None,
vocoder_config_path: str = None,
*,
model_path: Optional[str] = None,
config_path: Optional[str] = None,
vocoder_name: Optional[str] = None,
vocoder_path: Optional[str] = None,
vocoder_config_path: Optional[str] = None,
encoder_path: Optional[str] = None,
encoder_config_path: Optional[str] = None,
speakers_file_path: Optional[str] = None,
language_ids_file_path: Optional[str] = None,
progress_bar: bool = True,
gpu=False,
):
gpu: bool = False,
) -> None:
"""🐸TTS python interface that allows to load and use the released models.
Example with a multi-speaker model:
@ -34,118 +43,147 @@ class TTS(nn.Module):
>>> tts.tts_to_file(text="Hello world!", speaker=tts.speakers[0], language=tts.languages[0], file_path="output.wav")
Example with a single-speaker model:
>>> tts = TTS(model_name="tts_models/de/thorsten/tacotron2-DDC", progress_bar=False, gpu=False)
>>> tts = TTS(model_name="tts_models/de/thorsten/tacotron2-DDC", progress_bar=False)
>>> tts.tts_to_file(text="Ich bin eine Testnachricht.", file_path="output.wav")
Example loading a model from a path:
>>> tts = TTS(model_path="/path/to/checkpoint_100000.pth", config_path="/path/to/config.json", progress_bar=False, gpu=False)
>>> tts = TTS(model_path="/path/to/checkpoint_100000.pth", config_path="/path/to/config.json", progress_bar=False)
>>> tts.tts_to_file(text="Ich bin eine Testnachricht.", file_path="output.wav")
Example voice cloning with YourTTS in English, French and Portuguese:
>>> tts = TTS(model_name="tts_models/multilingual/multi-dataset/your_tts", progress_bar=False, gpu=True)
>>> tts = TTS(model_name="tts_models/multilingual/multi-dataset/your_tts", progress_bar=False).to("cuda")
>>> tts.tts_to_file("This is voice cloning.", speaker_wav="my/cloning/audio.wav", language="en", file_path="thisisit.wav")
>>> tts.tts_to_file("C'est le clonage de la voix.", speaker_wav="my/cloning/audio.wav", language="fr", file_path="thisisit.wav")
>>> tts.tts_to_file("Isso é clonagem de voz.", speaker_wav="my/cloning/audio.wav", language="pt", file_path="thisisit.wav")
Example Fairseq TTS models (uses ISO language codes in https://dl.fbaipublicfiles.com/mms/tts/all-tts-languages.html):
>>> tts = TTS(model_name="tts_models/eng/fairseq/vits", progress_bar=False, gpu=True)
>>> tts = TTS(model_name="tts_models/eng/fairseq/vits", progress_bar=False).to("cuda")
>>> tts.tts_to_file("This is a test.", file_path="output.wav")
Args:
model_name (str, optional): Model name to load. You can list models by ```tts.models```. Defaults to None.
model_path (str, optional): Path to the model checkpoint. Defaults to None.
config_path (str, optional): Path to the model config. Defaults to None.
vocoder_name (str, optional): Pre-trained vocoder to use. Defaults to None, i.e. using the default vocoder.
vocoder_path (str, optional): Path to the vocoder checkpoint. Defaults to None.
vocoder_config_path (str, optional): Path to the vocoder config. Defaults to None.
progress_bar (bool, optional): Whether to pring a progress bar while downloading a model. Defaults to True.
gpu (bool, optional): Enable/disable GPU. Some models might be too slow on CPU. Defaults to False.
encoder_path: Path to speaker encoder checkpoint. Default to None.
encoder_config_path: Path to speaker encoder config file. Defaults to None.
speakers_file_path: JSON file for multi-speaker model. Defaults to None.
language_ids_file_path: JSON file for multilingual model. Defaults to None
progress_bar (bool, optional): Whether to print a progress bar while downloading a model. Defaults to True.
gpu (bool, optional): Enable/disable GPU. Defaults to False. DEPRECATED, use TTS(...).to("cuda")
"""
super().__init__()
self.manager = ModelManager(models_file=self.get_models_file_path(), progress_bar=progress_bar, verbose=False)
self.manager = ModelManager(models_file=self.get_models_file_path(), progress_bar=progress_bar)
self.config = load_config(config_path) if config_path else None
self.synthesizer = None
self.voice_converter = None
self.model_name = ""
self.vocoder_path = vocoder_path
self.vocoder_config_path = vocoder_config_path
self.encoder_path = encoder_path
self.encoder_config_path = encoder_config_path
self.speakers_file_path = speakers_file_path
self.language_ids_file_path = language_ids_file_path
if gpu:
warnings.warn("`gpu` will be deprecated. Please use `tts.to(device)` instead.")
if model_name is not None and len(model_name) > 0:
if "tts_models" in model_name:
self.load_tts_model_by_name(model_name, gpu)
self.load_tts_model_by_name(model_name, vocoder_name, gpu=gpu)
elif "voice_conversion_models" in model_name:
self.load_vc_model_by_name(model_name, gpu)
self.load_vc_model_by_name(model_name, gpu=gpu)
# To allow just TTS("xtts")
else:
self.load_model_by_name(model_name, gpu)
self.load_model_by_name(model_name, vocoder_name, gpu=gpu)
if model_path:
self.load_tts_model_by_path(
model_path, config_path, vocoder_path=vocoder_path, vocoder_config=vocoder_config_path, gpu=gpu
)
self.load_tts_model_by_path(model_path, config_path, gpu=gpu)
@property
def models(self):
def models(self) -> list[str]:
return self.manager.list_tts_models()
@property
def is_multi_speaker(self):
if hasattr(self.synthesizer.tts_model, "speaker_manager") and self.synthesizer.tts_model.speaker_manager:
def is_multi_speaker(self) -> bool:
if (
self.synthesizer is not None
and hasattr(self.synthesizer.tts_model, "speaker_manager")
and self.synthesizer.tts_model.speaker_manager
):
return self.synthesizer.tts_model.speaker_manager.num_speakers > 1
return False
@property
def is_multi_lingual(self):
def is_multi_lingual(self) -> bool:
# Not sure what sets this to None, but applied a fix to prevent crashing.
if (
isinstance(self.model_name, str)
and "xtts" in self.model_name
or self.config
and ("xtts" in self.config.model or len(self.config.languages) > 1)
and ("xtts" in self.config.model or "languages" in self.config and len(self.config.languages) > 1)
):
return True
if hasattr(self.synthesizer.tts_model, "language_manager") and self.synthesizer.tts_model.language_manager:
if (
self.synthesizer is not None
and hasattr(self.synthesizer.tts_model, "language_manager")
and self.synthesizer.tts_model.language_manager
):
return self.synthesizer.tts_model.language_manager.num_languages > 1
return False
@property
def speakers(self):
def speakers(self) -> list[str]:
if not self.is_multi_speaker:
return None
return self.synthesizer.tts_model.speaker_manager.speaker_names
@property
def languages(self):
def languages(self) -> list[str]:
if not self.is_multi_lingual:
return None
return self.synthesizer.tts_model.language_manager.language_names
@staticmethod
def get_models_file_path():
def get_models_file_path() -> Path:
return Path(__file__).parent / ".models.json"
def list_models(self):
return ModelManager(models_file=TTS.get_models_file_path(), progress_bar=False, verbose=False)
@staticmethod
def list_models() -> list[str]:
return ModelManager(models_file=TTS.get_models_file_path(), progress_bar=False).list_models()
def download_model_by_name(self, model_name: str):
def download_model_by_name(
self, model_name: str, vocoder_name: Optional[str] = None
) -> tuple[Optional[Path], Optional[Path], Optional[Path]]:
model_path, config_path, model_item = self.manager.download_model(model_name)
if "fairseq" in model_name or (model_item is not None and isinstance(model_item["model_url"], list)):
# return model directory if there are multiple files
# we assume that the model knows how to load itself
return None, None, None, None, model_path
return None, None, model_path
if model_item.get("default_vocoder") is None:
return model_path, config_path, None, None, None
vocoder_path, vocoder_config_path, _ = self.manager.download_model(model_item["default_vocoder"])
return model_path, config_path, vocoder_path, vocoder_config_path, None
return model_path, config_path, None
if vocoder_name is None:
vocoder_name = model_item["default_vocoder"]
vocoder_path, vocoder_config_path, _ = self.manager.download_model(vocoder_name)
# A local vocoder model will take precedence if specified via vocoder_path
if self.vocoder_path is None or self.vocoder_config_path is None:
self.vocoder_path = vocoder_path
self.vocoder_config_path = vocoder_config_path
return model_path, config_path, None
def load_model_by_name(self, model_name: str, gpu: bool = False):
def load_model_by_name(self, model_name: str, vocoder_name: Optional[str] = None, *, gpu: bool = False) -> None:
"""Load one of the 🐸TTS models by name.
Args:
model_name (str): Model name to load. You can list models by ```tts.models```.
gpu (bool, optional): Enable/disable GPU. Some models might be too slow on CPU. Defaults to False.
"""
self.load_tts_model_by_name(model_name, gpu)
self.load_tts_model_by_name(model_name, vocoder_name, gpu=gpu)
def load_vc_model_by_name(self, model_name: str, gpu: bool = False):
def load_vc_model_by_name(self, model_name: str, *, gpu: bool = False) -> None:
"""Load one of the voice conversion models by name.
Args:
@ -153,10 +191,12 @@ class TTS(nn.Module):
gpu (bool, optional): Enable/disable GPU. Some models might be too slow on CPU. Defaults to False.
"""
self.model_name = model_name
model_path, config_path, _, _, _ = self.download_model_by_name(model_name)
self.voice_converter = Synthesizer(vc_checkpoint=model_path, vc_config=config_path, use_cuda=gpu)
model_path, config_path, model_dir = self.download_model_by_name(model_name)
self.voice_converter = Synthesizer(
vc_checkpoint=model_path, vc_config=config_path, model_dir=model_dir, use_cuda=gpu
)
def load_tts_model_by_name(self, model_name: str, gpu: bool = False):
def load_tts_model_by_name(self, model_name: str, vocoder_name: Optional[str] = None, *, gpu: bool = False) -> None:
"""Load one of 🐸TTS models by name.
Args:
@ -168,9 +208,7 @@ class TTS(nn.Module):
self.synthesizer = None
self.model_name = model_name
model_path, config_path, vocoder_path, vocoder_config_path, model_dir = self.download_model_by_name(
model_name
)
model_path, config_path, model_dir = self.download_model_by_name(model_name, vocoder_name)
# init synthesizer
# None values are fetch from the model
@ -179,17 +217,15 @@ class TTS(nn.Module):
tts_config_path=config_path,
tts_speakers_file=None,
tts_languages_file=None,
vocoder_checkpoint=vocoder_path,
vocoder_config=vocoder_config_path,
encoder_checkpoint=None,
encoder_config=None,
vocoder_checkpoint=self.vocoder_path,
vocoder_config=self.vocoder_config_path,
encoder_checkpoint=self.encoder_path,
encoder_config=self.encoder_config_path,
model_dir=model_dir,
use_cuda=gpu,
)
def load_tts_model_by_path(
self, model_path: str, config_path: str, vocoder_path: str = None, vocoder_config: str = None, gpu: bool = False
):
def load_tts_model_by_path(self, model_path: str, config_path: str, *, gpu: bool = False) -> None:
"""Load a model from a path.
Args:
@ -203,22 +239,22 @@ class TTS(nn.Module):
self.synthesizer = Synthesizer(
tts_checkpoint=model_path,
tts_config_path=config_path,
tts_speakers_file=None,
tts_languages_file=None,
vocoder_checkpoint=vocoder_path,
vocoder_config=vocoder_config,
encoder_checkpoint=None,
encoder_config=None,
tts_speakers_file=self.speakers_file_path,
tts_languages_file=self.language_ids_file_path,
vocoder_checkpoint=self.vocoder_path,
vocoder_config=self.vocoder_config_path,
encoder_checkpoint=self.encoder_path,
encoder_config=self.encoder_config_path,
use_cuda=gpu,
)
def _check_arguments(
self,
speaker: str = None,
language: str = None,
speaker_wav: str = None,
emotion: str = None,
speed: float = None,
speaker: Optional[str] = None,
language: Optional[str] = None,
speaker_wav: Optional[str] = None,
emotion: Optional[str] = None,
speed: Optional[float] = None,
**kwargs,
) -> None:
"""Check if the arguments are valid for the model."""
@ -231,7 +267,7 @@ class TTS(nn.Module):
raise ValueError("Model is not multi-speaker but `speaker` is provided.")
if not self.is_multi_lingual and language is not None:
raise ValueError("Model is not multi-lingual but `language` is provided.")
if not emotion is None and not speed is None:
if emotion is not None and speed is not None:
raise ValueError("Emotion and speed can only be used with Coqui Studio models. Which is discontinued.")
def tts(
@ -278,10 +314,6 @@ class TTS(nn.Module):
speaker_name=speaker,
language_name=language,
speaker_wav=speaker_wav,
reference_wav=None,
style_wav=None,
style_text=None,
reference_speaker_name=None,
split_sentences=split_sentences,
speed=speed,
**kwargs,
@ -300,7 +332,7 @@ class TTS(nn.Module):
file_path: str = "output.wav",
split_sentences: bool = True,
**kwargs,
):
) -> str:
"""Convert text to speech.
Args:
@ -356,15 +388,18 @@ class TTS(nn.Module):
target_wav (str):`
Path to the target wav file.
"""
wav = self.voice_converter.voice_conversion(source_wav=source_wav, target_wav=target_wav)
return wav
if self.voice_converter is None:
msg = "The selected model does not support voice conversion."
raise RuntimeError(msg)
return self.voice_converter.voice_conversion(source_wav=source_wav, target_wav=target_wav)
def voice_conversion_to_file(
self,
source_wav: str,
target_wav: str,
file_path: str = "output.wav",
):
pipe_out=None,
) -> str:
"""Voice conversion with FreeVC. Convert source wav to target speaker.
Args:
@ -374,9 +409,11 @@ class TTS(nn.Module):
Path to the target wav file.
file_path (str, optional):
Output file path. Defaults to "output.wav".
pipe_out (BytesIO, optional):
Flag to stdout the generated TTS wav file for shell pipe.
"""
wav = self.voice_conversion(source_wav=source_wav, target_wav=target_wav)
save_wav(wav=wav, path=file_path, sample_rate=self.voice_converter.vc_config.audio.output_sample_rate)
self.voice_converter.save_wav(wav=wav, path=file_path, pipe_out=pipe_out)
return file_path
def tts_with_vc(
@ -429,7 +466,8 @@ class TTS(nn.Module):
file_path: str = "output.wav",
speaker: str = None,
split_sentences: bool = True,
):
pipe_out=None,
) -> str:
"""Convert text to speech with voice conversion and save to file.
Check `tts_with_vc` for more details.
@ -452,8 +490,11 @@ class TTS(nn.Module):
Split text into sentences, synthesize them separately and concatenate the file audio.
Setting it False uses more VRAM and possibly hit model specific text length or VRAM limits. Only
applicable to the 🐸TTS models. Defaults to True.
pipe_out (BytesIO, optional):
Flag to stdout the generated TTS wav file for shell pipe.
"""
wav = self.tts_with_vc(
text=text, language=language, speaker_wav=speaker_wav, speaker=speaker, split_sentences=split_sentences
)
save_wav(wav=wav, path=file_path, sample_rate=self.voice_converter.vc_config.audio.output_sample_rate)
self.voice_converter.save_wav(wav=wav, path=file_path, pipe_out=pipe_out)
return file_path

7
TTS/bin/collect_env_info.py
View File

@ -1,4 +1,6 @@
"""Get detailed info about the working environment."""
import json
import os
import platform
import sys
@ -6,11 +8,10 @@ import sys
import numpy
import torch
sys.path += [os.path.abspath(".."), os.path.abspath(".")]
import json
import TTS
sys.path += [os.path.abspath(".."), os.path.abspath(".")]
def system_info():
return {

15
TTS/bin/compute_attention_masks.py
View File

@ -1,21 +1,26 @@
import argparse
import importlib
import logging
import os
import sys
from argparse import RawTextHelpFormatter
import numpy as np
import torch
from torch.utils.data import DataLoader
from tqdm import tqdm
from trainer.io import load_checkpoint
from TTS.config import load_config
from TTS.tts.datasets.TTSDataset import TTSDataset
from TTS.tts.models import setup_model
from TTS.tts.utils.text.characters import make_symbols, phonemes, symbols
from TTS.utils.audio import AudioProcessor
from TTS.utils.io import load_checkpoint
from TTS.utils.generic_utils import ConsoleFormatter, setup_logger
if __name__ == "__main__":
setup_logger("TTS", level=logging.INFO, stream=sys.stdout, formatter=ConsoleFormatter())
# pylint: disable=bad-option-value
parser = argparse.ArgumentParser(
description="""Extract attention masks from trained Tacotron/Tacotron2 models.
@ -31,7 +36,7 @@ Example run:
--data_path /root/LJSpeech-1.1/
--batch_size 32
--dataset ljspeech
--use_cuda True
--use_cuda
""",
formatter_class=RawTextHelpFormatter,
)
@ -58,7 +63,7 @@ Example run:
help="Dataset metafile inclusing file paths with transcripts.",
)
parser.add_argument("--data_path", type=str, default="", help="Defines the data path. It overwrites config.json.")
parser.add_argument("--use_cuda", type=bool, default=False, help="enable/disable cuda.")
parser.add_argument("--use_cuda", action=argparse.BooleanOptionalAction, default=False, help="enable/disable cuda.")
parser.add_argument(
"--batch_size", default=16, type=int, help="Batch size for the model. Use batch_size=1 if you have no CUDA."
@ -70,13 +75,13 @@ Example run:
# if the vocabulary was passed, replace the default
if "characters" in C.keys():
symbols, phonemes = make_symbols(**C.characters)
symbols, phonemes = make_symbols(**C.characters) # noqa: F811
# load the model
num_chars = len(phonemes) if C.use_phonemes else len(symbols)
# TODO: handle multi-speaker
model = setup_model(C)
model, _ = load_checkpoint(model, args.model_path, args.use_cuda, True)
model, _ = load_checkpoint(model, args.model_path, use_cuda=args.use_cuda, eval=True)
# data loader
preprocessor = importlib.import_module("TTS.tts.datasets.formatters")

7
TTS/bin/compute_embeddings.py
View File

@ -1,5 +1,7 @@
import argparse
import logging
import os
import sys
from argparse import RawTextHelpFormatter
import torch
@ -10,6 +12,7 @@ from TTS.config.shared_configs import BaseDatasetConfig
from TTS.tts.datasets import load_tts_samples
from TTS.tts.utils.managers import save_file
from TTS.tts.utils.speakers import SpeakerManager
from TTS.utils.generic_utils import ConsoleFormatter, setup_logger
def compute_embeddings(
@ -100,6 +103,8 @@ def compute_embeddings(
if __name__ == "__main__":
setup_logger("TTS", level=logging.INFO, stream=sys.stdout, formatter=ConsoleFormatter())
parser = argparse.ArgumentParser(
description="""Compute embedding vectors for each audio file in a dataset and store them keyed by `{dataset_name}#{file_path}` in a .pth file\n\n"""
"""
@ -146,7 +151,7 @@ if __name__ == "__main__":
default=False,
action="store_true",
)
parser.add_argument("--disable_cuda", type=bool, help="Flag to disable cuda.", default=False)
parser.add_argument("--disable_cuda", action="store_true", help="Flag to disable cuda.", default=False)
parser.add_argument("--no_eval", help="Do not compute eval?. Default False", default=False, action="store_true")
parser.add_argument(
"--formatter_name",

5
TTS/bin/compute_statistics.py
View File

@ -3,7 +3,9 @@
import argparse
import glob
import logging
import os
import sys
import numpy as np
from tqdm import tqdm
@ -12,10 +14,13 @@ from tqdm import tqdm
from TTS.config import load_config
from TTS.tts.datasets import load_tts_samples
from TTS.utils.audio import AudioProcessor
from TTS.utils.generic_utils import ConsoleFormatter, setup_logger
def main():
"""Run preprocessing process."""
setup_logger("TTS", level=logging.INFO, stream=sys.stderr, formatter=ConsoleFormatter())
parser = argparse.ArgumentParser(description="Compute mean and variance of spectrogtram features.")
parser.add_argument("config_path", type=str, help="TTS config file path to define audio processin parameters.")
parser.add_argument("out_path", type=str, help="save path (directory and filename).")

9
TTS/bin/eval_encoder.py
View File

@ -1,4 +1,6 @@
import argparse
import logging
import sys
from argparse import RawTextHelpFormatter
import torch
@ -7,6 +9,7 @@ from tqdm import tqdm
from TTS.config import load_config
from TTS.tts.datasets import load_tts_samples
from TTS.tts.utils.speakers import SpeakerManager
from TTS.utils.generic_utils import ConsoleFormatter, setup_logger
def compute_encoder_accuracy(dataset_items, encoder_manager):
@ -51,6 +54,8 @@ def compute_encoder_accuracy(dataset_items, encoder_manager):
if __name__ == "__main__":
setup_logger("TTS", level=logging.INFO, stream=sys.stdout, formatter=ConsoleFormatter())
parser = argparse.ArgumentParser(
description="""Compute the accuracy of the encoder.\n\n"""
"""
@ -71,8 +76,8 @@ if __name__ == "__main__":
type=str,
help="Path to dataset config file.",
)
parser.add_argument("--use_cuda", type=bool, help="flag to set cuda.", default=True)
parser.add_argument("--eval", type=bool, help="compute eval.", default=True)
parser.add_argument("--use_cuda", action=argparse.BooleanOptionalAction, help="flag to set cuda.", default=True)
parser.add_argument("--eval", action=argparse.BooleanOptionalAction, help="compute eval.", default=True)
args = parser.parse_args()

14
TTS/bin/extract_tts_spectrograms.py
View File

@ -2,12 +2,15 @@
"""Extract Mel spectrograms with teacher forcing."""
import argparse
import logging
import os
import sys
import numpy as np
import torch
from torch.utils.data import DataLoader
from tqdm import tqdm
from trainer.generic_utils import count_parameters
from TTS.config import load_config
from TTS.tts.datasets import TTSDataset, load_tts_samples
@ -16,12 +19,12 @@ from TTS.tts.utils.speakers import SpeakerManager
from TTS.tts.utils.text.tokenizer import TTSTokenizer
from TTS.utils.audio import AudioProcessor
from TTS.utils.audio.numpy_transforms import quantize
from TTS.utils.generic_utils import count_parameters
from TTS.utils.generic_utils import ConsoleFormatter, setup_logger
use_cuda = torch.cuda.is_available()
def setup_loader(ap, r, verbose=False):
def setup_loader(ap, r):
tokenizer, _ = TTSTokenizer.init_from_config(c)
dataset = TTSDataset(
outputs_per_step=r,
@ -37,7 +40,6 @@ def setup_loader(ap, r, verbose=False):
phoneme_cache_path=c.phoneme_cache_path,
precompute_num_workers=0,
use_noise_augment=False,
verbose=verbose,
speaker_id_mapping=speaker_manager.name_to_id if c.use_speaker_embedding else None,
d_vector_mapping=speaker_manager.embeddings if c.use_d_vector_file else None,
)
@ -257,7 +259,7 @@ def main(args): # pylint: disable=redefined-outer-name
print("\n > Model has {} parameters".format(num_params), flush=True)
# set r
r = 1 if c.model.lower() == "glow_tts" else model.decoder.r
own_loader = setup_loader(ap, r, verbose=True)
own_loader = setup_loader(ap, r)
extract_spectrograms(
own_loader,
@ -272,6 +274,8 @@ def main(args): # pylint: disable=redefined-outer-name
if __name__ == "__main__":
setup_logger("TTS", level=logging.INFO, stream=sys.stdout, formatter=ConsoleFormatter())
parser = argparse.ArgumentParser()
parser.add_argument("--config_path", type=str, help="Path to config file for training.", required=True)
parser.add_argument("--checkpoint_path", type=str, help="Model file to be restored.", required=True)
@ -279,7 +283,7 @@ if __name__ == "__main__":
parser.add_argument("--debug", default=False, action="store_true", help="Save audio files for debug")
parser.add_argument("--save_audio", default=False, action="store_true", help="Save audio files")
parser.add_argument("--quantize_bits", type=int, default=0, help="Save quantized audio files if non-zero")
parser.add_argument("--eval", type=bool, help="compute eval.", default=True)
parser.add_argument("--eval", action=argparse.BooleanOptionalAction, help="compute eval.", default=True)
args = parser.parse_args()
c = load_config(args.config_path)

20
TTS/bin/find_unique_chars.py
View File

@ -1,12 +1,18 @@
"""Find all the unique characters in a dataset"""
import argparse
import logging
import sys
from argparse import RawTextHelpFormatter
from TTS.config import load_config
from TTS.tts.datasets import load_tts_samples
from TTS.tts.datasets import find_unique_chars, load_tts_samples
from TTS.utils.generic_utils import ConsoleFormatter, setup_logger
def main():
setup_logger("TTS", level=logging.INFO, stream=sys.stdout, formatter=ConsoleFormatter())
# pylint: disable=bad-option-value
parser = argparse.ArgumentParser(
description="""Find all the unique characters or phonemes in a dataset.\n\n"""
@ -28,17 +34,7 @@ def main():
)
items = train_items + eval_items
texts = "".join(item["text"] for item in items)
chars = set(texts)
lower_chars = filter(lambda c: c.islower(), chars)
chars_force_lower = [c.lower() for c in chars]
chars_force_lower = set(chars_force_lower)
print(f" > Number of unique characters: {len(chars)}")
print(f" > Unique characters: {''.join(sorted(chars))}")
print(f" > Unique lower characters: {''.join(sorted(lower_chars))}")
print(f" > Unique all forced to lower characters: {''.join(sorted(chars_force_lower))}")
find_unique_chars(items)
if __name__ == "__main__":

8
TTS/bin/find_unique_phonemes.py
View File

@ -1,6 +1,9 @@
"""Find all the unique characters in a dataset"""
import argparse
import logging
import multiprocessing
import sys
from argparse import RawTextHelpFormatter
from tqdm.contrib.concurrent import process_map
@ -8,15 +11,18 @@ from tqdm.contrib.concurrent import process_map
from TTS.config import load_config
from TTS.tts.datasets import load_tts_samples
from TTS.tts.utils.text.phonemizers import Gruut
from TTS.utils.generic_utils import ConsoleFormatter, setup_logger
def compute_phonemes(item):
text = item["text"]
ph = phonemizer.phonemize(text).replace("|", "")
return set(list(ph))
return set(ph)
def main():
setup_logger("TTS", level=logging.INFO, stream=sys.stdout, formatter=ConsoleFormatter())
# pylint: disable=W0601
global c, phonemizer
# pylint: disable=bad-option-value

15
TTS/bin/remove_silence_using_vad.py
View File

@ -1,12 +1,15 @@
import argparse
import glob
import logging
import multiprocessing
import os
import pathlib
import sys
import torch
from tqdm import tqdm
from TTS.utils.generic_utils import ConsoleFormatter, setup_logger
from TTS.utils.vad import get_vad_model_and_utils, remove_silence
torch.set_num_threads(1)
@ -75,8 +78,10 @@ def preprocess_audios():
if __name__ == "__main__":
setup_logger("TTS", level=logging.INFO, stream=sys.stdout, formatter=ConsoleFormatter())
parser = argparse.ArgumentParser(
description="python TTS/bin/remove_silence_using_vad.py -i=VCTK-Corpus/ -o=VCTK-Corpus-removed-silence/ -g=wav48_silence_trimmed/*/*_mic1.flac --trim_just_beginning_and_end True"
description="python TTS/bin/remove_silence_using_vad.py -i=VCTK-Corpus/ -o=VCTK-Corpus-removed-silence/ -g=wav48_silence_trimmed/*/*_mic1.flac --trim_just_beginning_and_end"
)
parser.add_argument("-i", "--input_dir", type=str, help="Dataset root dir", required=True)
parser.add_argument("-o", "--output_dir", type=str, help="Output Dataset dir", default="")
@ -91,20 +96,20 @@ if __name__ == "__main__":
parser.add_argument(
"-t",
"--trim_just_beginning_and_end",
type=bool,
action=argparse.BooleanOptionalAction,
default=True,
help="If True this script will trim just the beginning and end nonspeech parts. If False all nonspeech parts will be trim. Default True",
help="If True this script will trim just the beginning and end nonspeech parts. If False all nonspeech parts will be trimmed.",
)
parser.add_argument(
"-c",
"--use_cuda",
type=bool,
action=argparse.BooleanOptionalAction,
default=False,
help="If True use cuda",
)
parser.add_argument(
"--use_onnx",
type=bool,
action=argparse.BooleanOptionalAction,
default=False,
help="If True use onnx",
)

324
TTS/bin/synthesize.py
View File

@ -1,147 +1,141 @@
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""Command line interface."""
import argparse
import contextlib
import logging
import sys
from argparse import RawTextHelpFormatter
# pylint: disable=redefined-outer-name, unused-argument
from pathlib import Path
from TTS.utils.generic_utils import ConsoleFormatter, setup_logger
logger = logging.getLogger(__name__)
description = """
Synthesize speech on command line.
Synthesize speech on the command line.
You can either use your trained model or choose a model from the provided list.
If you don't specify any models, then it uses LJSpeech based English model.
#### Single Speaker Models
- List provided models:
```
$ tts --list_models
```sh
tts --list_models
```
- Get model info (for both tts_models and vocoder_models):
- Query by type/name:
The model_info_by_name uses the name as it from the --list_models.
```
$ tts --model_info_by_name "<model_type>/<language>/<dataset>/<model_name>"
- Get model information. Use the names obtained from `--list_models`.
```sh
tts --model_info_by_name "<model_type>/<language>/<dataset>/<model_name>"
```
For example:
```
$ tts --model_info_by_name tts_models/tr/common-voice/glow-tts
$ tts --model_info_by_name vocoder_models/en/ljspeech/hifigan_v2
```
- Query by type/idx:
The model_query_idx uses the corresponding idx from --list_models.
```
$ tts --model_info_by_idx "<model_type>/<model_query_idx>"
```sh
tts --model_info_by_name tts_models/tr/common-voice/glow-tts
tts --model_info_by_name vocoder_models/en/ljspeech/hifigan_v2
```
For example:
#### Single speaker models
```
$ tts --model_info_by_idx tts_models/3
```
- Run TTS with the default model (`tts_models/en/ljspeech/tacotron2-DDC`):
- Query info for model info by full name:
```
$ tts --model_info_by_name "<model_type>/<language>/<dataset>/<model_name>"
```
- Run TTS with default models:
```
$ tts --text "Text for TTS" --out_path output/path/speech.wav
```sh
tts --text "Text for TTS" --out_path output/path/speech.wav
```
- Run TTS and pipe out the generated TTS wav file data:
```
$ tts --text "Text for TTS" --pipe_out --out_path output/path/speech.wav | aplay
```sh
tts --text "Text for TTS" --pipe_out --out_path output/path/speech.wav | aplay
```
- Run a TTS model with its default vocoder model:
```
$ tts --text "Text for TTS" --model_name "<model_type>/<language>/<dataset>/<model_name>" --out_path output/path/speech.wav
```sh
tts --text "Text for TTS" \\
--model_name "<model_type>/<language>/<dataset>/<model_name>" \\
--out_path output/path/speech.wav
```
For example:
```
$ tts --text "Text for TTS" --model_name "tts_models/en/ljspeech/glow-tts" --out_path output/path/speech.wav
```sh
tts --text "Text for TTS" \\
--model_name "tts_models/en/ljspeech/glow-tts" \\
--out_path output/path/speech.wav
```
- Run with specific TTS and vocoder models from the list:
- Run with specific TTS and vocoder models from the list. Note that not every vocoder is compatible with every TTS model.
```
$ tts --text "Text for TTS" --model_name "<model_type>/<language>/<dataset>/<model_name>" --vocoder_name "<model_type>/<language>/<dataset>/<model_name>" --out_path output/path/speech.wav
```sh
tts --text "Text for TTS" \\
--model_name "<model_type>/<language>/<dataset>/<model_name>" \\
--vocoder_name "<model_type>/<language>/<dataset>/<model_name>" \\
--out_path output/path/speech.wav
```
For example:
```
$ tts --text "Text for TTS" --model_name "tts_models/en/ljspeech/glow-tts" --vocoder_name "vocoder_models/en/ljspeech/univnet" --out_path output/path/speech.wav
```sh
tts --text "Text for TTS" \\
--model_name "tts_models/en/ljspeech/glow-tts" \\
--vocoder_name "vocoder_models/en/ljspeech/univnet" \\
--out_path output/path/speech.wav
```
- Run your own TTS model (Using Griffin-Lim Vocoder):
- Run your own TTS model (using Griffin-Lim Vocoder):
```
$ tts --text "Text for TTS" --model_path path/to/model.pth --config_path path/to/config.json --out_path output/path/speech.wav
```sh
tts --text "Text for TTS" \\
--model_path path/to/model.pth \\
--config_path path/to/config.json \\
--out_path output/path/speech.wav
```
- Run your own TTS and Vocoder models:
```
$ tts --text "Text for TTS" --model_path path/to/model.pth --config_path path/to/config.json --out_path output/path/speech.wav
--vocoder_path path/to/vocoder.pth --vocoder_config_path path/to/vocoder_config.json
```sh
tts --text "Text for TTS" \\
--model_path path/to/model.pth \\
--config_path path/to/config.json \\
--out_path output/path/speech.wav \\
--vocoder_path path/to/vocoder.pth \\
--vocoder_config_path path/to/vocoder_config.json
```
#### Multi-speaker Models
#### Multi-speaker models
- List the available speakers and choose a <speaker_id> among them:
- List the available speakers and choose a `<speaker_id>` among them:
```
$ tts --model_name "<language>/<dataset>/<model_name>" --list_speaker_idxs
```sh
tts --model_name "<language>/<dataset>/<model_name>" --list_speaker_idxs
```
- Run the multi-speaker TTS model with the target speaker ID:
```
$ tts --text "Text for TTS." --out_path output/path/speech.wav --model_name "<language>/<dataset>/<model_name>" --speaker_idx <speaker_id>
```sh
tts --text "Text for TTS." --out_path output/path/speech.wav \\
--model_name "<language>/<dataset>/<model_name>" --speaker_idx <speaker_id>
```
- Run your own multi-speaker TTS model:
```
$ tts --text "Text for TTS" --out_path output/path/speech.wav --model_path path/to/model.pth --config_path path/to/config.json --speakers_file_path path/to/speaker.json --speaker_idx <speaker_id>
```sh
tts --text "Text for TTS" --out_path output/path/speech.wav \\
--model_path path/to/model.pth --config_path path/to/config.json \\
--speakers_file_path path/to/speaker.json --speaker_idx <speaker_id>
```
### Voice Conversion Models
#### Voice conversion models
```
$ tts --out_path output/path/speech.wav --model_name "<language>/<dataset>/<model_name>" --source_wav <path/to/speaker/wav> --target_wav <path/to/reference/wav>
```sh
tts --out_path output/path/speech.wav --model_name "<language>/<dataset>/<model_name>" \\
--source_wav <path/to/speaker/wav> --target_wav <path/to/reference/wav>
```
"""
def str2bool(v):
if isinstance(v, bool):
return v
if v.lower() in ("yes", "true", "t", "y", "1"):
return True
if v.lower() in ("no", "false", "f", "n", "0"):
return False
raise argparse.ArgumentTypeError("Boolean value expected.")
def main():
def parse_args() -> argparse.Namespace:
"""Parse arguments."""
parser = argparse.ArgumentParser(
description=description.replace(" ```\n", ""),
formatter_class=RawTextHelpFormatter,
@ -149,10 +143,7 @@ def main():
parser.add_argument(
"--list_models",
type=str2bool,
nargs="?",
const=True,
default=False,
action="store_true",
help="list available pre-trained TTS and vocoder models.",
)
@ -200,7 +191,7 @@ def main():
default="tts_output.wav",
help="Output wav file path.",
)
parser.add_argument("--use_cuda", type=bool, help="Run model on CUDA.", default=False)
parser.add_argument("--use_cuda", action="store_true", help="Run model on CUDA.")
parser.add_argument("--device", type=str, help="Device to run model on.", default="cpu")
parser.add_argument(
"--vocoder_path",
@ -219,10 +210,7 @@ def main():
parser.add_argument(
"--pipe_out",
help="stdout the generated TTS wav file for shell pipe.",
type=str2bool,
nargs="?",
const=True,
default=False,
action="store_true",
)
# args for multi-speaker synthesis
@ -254,26 +242,14 @@ def main():
parser.add_argument(
"--list_speaker_idxs",
help="List available speaker ids for the defined multi-speaker model.",
type=str2bool,
nargs="?",
const=True,
default=False,
action="store_true",
)
parser.add_argument(
"--list_language_idxs",
help="List available language ids for the defined multi-lingual model.",
type=str2bool,
nargs="?",
const=True,
default=False,
action="store_true",
)
# aux args
parser.add_argument(
"--save_spectogram",
type=bool,
help="If true save raw spectogram for further (vocoder) processing in out_path.",
default=False,
)
parser.add_argument(
"--reference_wav",
type=str,
@ -288,8 +264,8 @@ def main():
)
parser.add_argument(
"--progress_bar",
type=str2bool,
help="If true shows a progress bar for the model download. Defaults to True",
action=argparse.BooleanOptionalAction,
help="Show a progress bar for the model download.",
default=True,
)
@ -330,6 +306,14 @@ def main():
]
if not any(check_args):
parser.parse_args(["-h"])
return args
def main() -> None:
"""Entry point for `tts` command line interface."""
args = parse_args()
stream = sys.stderr if args.pipe_out else sys.stdout
setup_logger("TTS", level=logging.INFO, stream=stream, formatter=ConsoleFormatter())
pipe_out = sys.stdout if args.pipe_out else None
@ -337,12 +321,9 @@ def main():
# Late-import to make things load faster
from TTS.api import TTS
from TTS.utils.manage import ModelManager
from TTS.utils.synthesizer import Synthesizer
# load model manager
path = Path(__file__).parent / "../.models.json"
manager = ModelManager(path, progress_bar=args.progress_bar)
api = TTS()
manager = ModelManager(models_file=TTS.get_models_file_path(), progress_bar=args.progress_bar)
tts_path = None
tts_config_path = None
@ -356,12 +337,12 @@ def main():
vc_config_path = None
model_dir = None
# CASE1 #list : list pre-trained TTS models
# 1) List pre-trained TTS models
if args.list_models:
manager.list_models()
sys.exit()
# CASE2 #info : model info for pre-trained TTS models
# 2) Info about pre-trained TTS models (without loading a model)
if args.model_info_by_idx:
model_query = args.model_info_by_idx
manager.model_info_by_idx(model_query)
@ -372,122 +353,83 @@ def main():
manager.model_info_by_full_name(model_query_full_name)
sys.exit()
# CASE3: load pre-trained model paths
if args.model_name is not None and not args.model_path:
model_path, config_path, model_item = manager.download_model(args.model_name)
# tts model
if model_item["model_type"] == "tts_models":
tts_path = model_path
tts_config_path = config_path
if "default_vocoder" in model_item:
args.vocoder_name = (
model_item["default_vocoder"] if args.vocoder_name is None else args.vocoder_name
)
# voice conversion model
if model_item["model_type"] == "voice_conversion_models":
vc_path = model_path
vc_config_path = config_path
# tts model with multiple files to be loaded from the directory path
if model_item.get("author", None) == "fairseq" or isinstance(model_item["model_url"], list):
model_dir = model_path
tts_path = None
tts_config_path = None
args.vocoder_name = None
# load vocoder
if args.vocoder_name is not None and not args.vocoder_path:
vocoder_path, vocoder_config_path, _ = manager.download_model(args.vocoder_name)
# CASE4: set custom model paths
if args.model_path is not None:
tts_path = args.model_path
tts_config_path = args.config_path
speakers_file_path = args.speakers_file_path
language_ids_file_path = args.language_ids_file_path
if args.vocoder_path is not None:
vocoder_path = args.vocoder_path
vocoder_config_path = args.vocoder_config_path
if args.encoder_path is not None:
encoder_path = args.encoder_path
encoder_config_path = args.encoder_config_path
# 3) Load a model for further info or TTS/VC
device = args.device
if args.use_cuda:
device = "cuda"
# load models
synthesizer = Synthesizer(
tts_path,
tts_config_path,
speakers_file_path,
language_ids_file_path,
vocoder_path,
vocoder_config_path,
encoder_path,
encoder_config_path,
vc_path,
vc_config_path,
model_dir,
args.voice_dir,
# A local model will take precedence if specified via modeL_path
model_name = args.model_name if args.model_path is None else None
api = TTS(
model_name=model_name,
model_path=args.model_path,
config_path=args.config_path,
vocoder_name=args.vocoder_name,
vocoder_path=args.vocoder_path,
vocoder_config_path=args.vocoder_config_path,
encoder_path=args.encoder_path,
encoder_config_path=args.encoder_config_path,
speakers_file_path=args.speakers_file_path,
language_ids_file_path=args.language_ids_file_path,
progress_bar=args.progress_bar,
).to(device)
# query speaker ids of a multi-speaker model.
if args.list_speaker_idxs:
print(
" > Available speaker ids: (Set --speaker_idx flag to one of these values to use the multi-speaker model."
if not api.is_multi_speaker:
logger.info("Model only has a single speaker.")
return
logger.info(
"Available speaker ids: (Set --speaker_idx flag to one of these values to use the multi-speaker model."
)
print(synthesizer.tts_model.speaker_manager.name_to_id)
logger.info(api.speakers)
return
# query langauge ids of a multi-lingual model.
if args.list_language_idxs:
print(
" > Available language ids: (Set --language_idx flag to one of these values to use the multi-lingual model."
if not api.is_multi_lingual:
logger.info("Monolingual model.")
return
logger.info(
"Available language ids: (Set --language_idx flag to one of these values to use the multi-lingual model."
)
print(synthesizer.tts_model.language_manager.name_to_id)
logger.info(api.languages)
return
# check the arguments against a multi-speaker model.
if synthesizer.tts_speakers_file and (not args.speaker_idx and not args.speaker_wav):
print(
" [!] Looks like you use a multi-speaker model. Define `--speaker_idx` to "
if api.is_multi_speaker and (not args.speaker_idx and not args.speaker_wav):
logger.error(
"Looks like you use a multi-speaker model. Define `--speaker_idx` to "
"select the target speaker. You can list the available speakers for this model by `--list_speaker_idxs`."
)
return
# RUN THE SYNTHESIS
if args.text:
print(" > Text: {}".format(args.text))
logger.info("Text: %s", args.text)
# kick it
if tts_path is not None:
wav = synthesizer.tts(
args.text,
speaker_name=args.speaker_idx,
language_name=args.language_idx,
if args.text is not None:
api.tts_to_file(
text=args.text,
speaker=args.speaker_idx,
language=args.language_idx,
speaker_wav=args.speaker_wav,
pipe_out=pipe_out,
file_path=args.out_path,
reference_wav=args.reference_wav,
style_wav=args.capacitron_style_wav,
style_text=args.capacitron_style_text,
reference_speaker_name=args.reference_speaker_idx,
voice_dir=args.voice_dir,
)
elif vc_path is not None:
wav = synthesizer.voice_conversion(
logger.info("Saved TTS output to %s", args.out_path)
elif args.source_wav is not None and args.target_wav is not None:
api.voice_conversion_to_file(
source_wav=args.source_wav,
target_wav=args.target_wav,
file_path=args.out_path,
pipe_out=pipe_out,
)
elif model_dir is not None:
wav = synthesizer.tts(
args.text, speaker_name=args.speaker_idx, language_name=args.language_idx, speaker_wav=args.speaker_wav
)
# save the results
print(" > Saving output to {}".format(args.out_path))
synthesizer.save_wav(wav, args.out_path, pipe_out=pipe_out)
logger.info("Saved VC output to %s", args.out_path)
if __name__ == "__main__":

24
TTS/bin/train_encoder.py
View File

@ -1,13 +1,16 @@
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
import logging
import os
import sys
import time
import traceback
import warnings
import torch
from torch.utils.data import DataLoader
from trainer.generic_utils import count_parameters, remove_experiment_folder
from trainer.io import copy_model_files, save_best_model, save_checkpoint
from trainer.torch import NoamLR
from trainer.trainer_utils import get_optimizer
@ -18,7 +21,7 @@ from TTS.encoder.utils.training import init_training
from TTS.encoder.utils.visual import plot_embeddings
from TTS.tts.datasets import load_tts_samples
from TTS.utils.audio import AudioProcessor
from TTS.utils.generic_utils import count_parameters, remove_experiment_folder
from TTS.utils.generic_utils import ConsoleFormatter, setup_logger
from TTS.utils.samplers import PerfectBatchSampler
from TTS.utils.training import check_update
@ -31,7 +34,7 @@ print(" > Using CUDA: ", use_cuda)
print(" > Number of GPUs: ", num_gpus)
def setup_loader(ap: AudioProcessor, is_val: bool = False, verbose: bool = False):
def setup_loader(ap: AudioProcessor, is_val: bool = False):
num_utter_per_class = c.num_utter_per_class if not is_val else c.eval_num_utter_per_class
num_classes_in_batch = c.num_classes_in_batch if not is_val else c.eval_num_classes_in_batch
@ -42,7 +45,6 @@ def setup_loader(ap: AudioProcessor, is_val: bool = False, verbose: bool = False
voice_len=c.voice_len,
num_utter_per_class=num_utter_per_class,
num_classes_in_batch=num_classes_in_batch,
verbose=verbose,
augmentation_config=c.audio_augmentation if not is_val else None,
use_torch_spec=c.model_params.get("use_torch_spec", False),
)
@ -115,11 +117,14 @@ def evaluation(model, criterion, data_loader, global_step):
eval_avg_loss = eval_loss / len(data_loader)
# save stats
dashboard_logger.eval_stats(global_step, {"loss": eval_avg_loss})
try:
# plot the last batch in the evaluation
figures = {
"UMAP Plot": plot_embeddings(outputs.detach().cpu().numpy(), c.num_classes_in_batch),
}
dashboard_logger.eval_figures(global_step, figures)
except ImportError:
warnings.warn("Install the `umap-learn` package to see embedding plots.")
return eval_avg_loss
@ -160,9 +165,6 @@ def train(model, optimizer, scheduler, criterion, data_loader, eval_data_loader,
loader_time = time.time() - end_time
global_step += 1
# setup lr
if c.lr_decay:
scheduler.step()
optimizer.zero_grad()
# dispatch data to GPU
@ -181,6 +183,10 @@ def train(model, optimizer, scheduler, criterion, data_loader, eval_data_loader,
grad_norm, _ = check_update(model, c.grad_clip)
optimizer.step()
# setup lr
if c.lr_decay:
scheduler.step()
step_time = time.time() - start_time
epoch_time += step_time
@ -278,9 +284,9 @@ def main(args): # pylint: disable=redefined-outer-name
# pylint: disable=redefined-outer-name
meta_data_train, meta_data_eval = load_tts_samples(c.datasets, eval_split=True)
train_data_loader, train_classes, map_classid_to_classname = setup_loader(ap, is_val=False, verbose=True)
train_data_loader, train_classes, map_classid_to_classname = setup_loader(ap, is_val=False)
if c.run_eval:
eval_data_loader, _, _ = setup_loader(ap, is_val=True, verbose=True)
eval_data_loader, _, _ = setup_loader(ap, is_val=True)
else:
eval_data_loader = None
@ -316,6 +322,8 @@ def main(args): # pylint: disable=redefined-outer-name
if __name__ == "__main__":
setup_logger("TTS", level=logging.INFO, stream=sys.stdout, formatter=ConsoleFormatter())
args, c, OUT_PATH, AUDIO_PATH, c_logger, dashboard_logger = init_training()
try:

5
TTS/bin/train_tts.py
View File

@ -1,4 +1,6 @@
import logging
import os
import sys
from dataclasses import dataclass, field
from trainer import Trainer, TrainerArgs
@ -6,6 +8,7 @@ from trainer import Trainer, TrainerArgs
from TTS.config import load_config, register_config
from TTS.tts.datasets import load_tts_samples
from TTS.tts.models import setup_model
from TTS.utils.generic_utils import ConsoleFormatter, setup_logger
@dataclass
@ -15,6 +18,8 @@ class TrainTTSArgs(TrainerArgs):
def main():
"""Run `tts` model training directly by a `config.json` file."""
setup_logger("TTS", level=logging.INFO, stream=sys.stdout, formatter=ConsoleFormatter())
# init trainer args
train_args = TrainTTSArgs()
parser = train_args.init_argparse(arg_prefix="")

5
TTS/bin/train_vocoder.py
View File

@ -1,10 +1,13 @@
import logging
import os
import sys
from dataclasses import dataclass, field
from trainer import Trainer, TrainerArgs
from TTS.config import load_config, register_config
from TTS.utils.audio import AudioProcessor
from TTS.utils.generic_utils import ConsoleFormatter, setup_logger
from TTS.vocoder.datasets.preprocess import load_wav_data, load_wav_feat_data
from TTS.vocoder.models import setup_model
@ -16,6 +19,8 @@ class TrainVocoderArgs(TrainerArgs):
def main():
"""Run `tts` model training directly by a `config.json` file."""
setup_logger("TTS", level=logging.INFO, stream=sys.stdout, formatter=ConsoleFormatter())
# init trainer args
train_args = TrainVocoderArgs()
parser = train_args.init_argparse(arg_prefix="")

7
TTS/bin/tune_wavegrad.py
View File

@ -1,5 +1,8 @@
"""Search a good noise schedule for WaveGrad for a given number of inference iterations"""
import argparse
import logging
import sys
from itertools import product as cartesian_product
import numpy as np
@ -9,11 +12,14 @@ from tqdm import tqdm
from TTS.config import load_config
from TTS.utils.audio import AudioProcessor
from TTS.utils.generic_utils import ConsoleFormatter, setup_logger
from TTS.vocoder.datasets.preprocess import load_wav_data
from TTS.vocoder.datasets.wavegrad_dataset import WaveGradDataset
from TTS.vocoder.models import setup_model
if __name__ == "__main__":
setup_logger("TTS", level=logging.INFO, stream=sys.stdout, formatter=ConsoleFormatter())
parser = argparse.ArgumentParser()
parser.add_argument("--model_path", type=str, help="Path to model checkpoint.")
parser.add_argument("--config_path", type=str, help="Path to model config file.")
@ -54,7 +60,6 @@ if __name__ == "__main__":
return_segments=False,
use_noise_augment=False,
use_cache=False,
verbose=True,
)
loader = DataLoader(
dataset,

10
TTS/config/__init__.py
View File

@ -1,7 +1,7 @@
import json
import os
import re
from typing import Dict
from typing import Any, Dict, Union
import fsspec
import yaml
@ -17,9 +17,12 @@ def read_json_with_comments(json_path):
with fsspec.open(json_path, "r", encoding="utf-8") as f:
input_str = f.read()
# handle comments but not urls with //
input_str = re.sub(r"(\"(?:[^\"\\]|\\.)*\")|(/\*(?:.|[\\n\\r])*?\*/)|(//.*)", lambda m: m.group(1) or m.group(2) or "", input_str)
input_str = re.sub(
r"(\"(?:[^\"\\]|\\.)*\")|(/\*(?:.|[\\n\\r])*?\*/)|(//.*)", lambda m: m.group(1) or m.group(2) or "", input_str
)
return json.loads(input_str)
def register_config(model_name: str) -> Coqpit:
"""Find the right config for the given model name.
@ -65,7 +68,7 @@ def _process_model_name(config_dict: Dict) -> str:
return model_name
def load_config(config_path: str) -> Coqpit:
def load_config(config_path: Union[str, os.PathLike[Any]]) -> Coqpit:
"""Import `json` or `yaml` files as TTS configs. First, load the input file as a `dict` and check the model name
to find the corresponding Config class. Then initialize the Config.
@ -78,6 +81,7 @@ def load_config(config_path: str) -> Coqpit:
Returns:
Coqpit: TTS config object.
"""
config_path = str(config_path)
config_dict = {}
ext = os.path.splitext(config_path)[1]
if ext in (".yml", ".yaml"):

51
TTS/demos/xtts_ft_demo/utils/formatter.py
View File

@ -1,23 +1,17 @@
import os
import gc
import torchaudio
import os
import pandas
from faster_whisper import WhisperModel
from glob import glob
from tqdm import tqdm
import torch
import torchaudio
# torch.set_num_threads(1)
from faster_whisper import WhisperModel
from tqdm import tqdm
# torch.set_num_threads(1)
from TTS.tts.layers.xtts.tokenizer import multilingual_cleaners
torch.set_num_threads(16)
import os
audio_types = (".wav", ".mp3", ".flac")
@ -25,9 +19,10 @@ def list_audios(basePath, contains=None):
# return the set of files that are valid
return list_files(basePath, validExts=audio_types, contains=contains)
def list_files(basePath, validExts=None, contains=None):
# loop over the directory structure
for (rootDir, dirNames, filenames) in os.walk(basePath):
for rootDir, dirNames, filenames in os.walk(basePath):
# loop over the filenames in the current directory
for filename in filenames:
# if the contains string is not none and the filename does not contain
@ -36,7 +31,7 @@ def list_files(basePath, validExts=None, contains=None):
continue
# determine the file extension of the current file
ext = filename[filename.rfind("."):].lower()
ext = filename[filename.rfind(".") :].lower()
# check to see if the file is an audio and should be processed
if validExts is None or ext.endswith(validExts):
@ -44,7 +39,16 @@ def list_files(basePath, validExts=None, contains=None):
audioPath = os.path.join(rootDir, filename)
yield audioPath
def format_audio_list(audio_files, target_language="en", out_path=None, buffer=0.2, eval_percentage=0.15, speaker_name="coqui", gradio_progress=None):
def format_audio_list(
audio_files,
target_language="en",
out_path=None,
buffer=0.2,
eval_percentage=0.15,
speaker_name="coqui",
gradio_progress=None,
):
audio_total_size = 0
# make sure that ooutput file exists
os.makedirs(out_path, exist_ok=True)
@ -69,7 +73,7 @@ def format_audio_list(audio_files, target_language="en", out_path=None, buffer=0
wav = torch.mean(wav, dim=0, keepdim=True)
wav = wav.squeeze()
audio_total_size += (wav.size(-1) / sr)
audio_total_size += wav.size(-1) / sr
segments, _ = asr_model.transcribe(audio_path, word_timestamps=True, language=target_language)
segments = list(segments)
@ -94,7 +98,7 @@ def format_audio_list(audio_files, target_language="en", out_path=None, buffer=0
# get previous sentence end
previous_word_end = words_list[word_idx - 1].end
# add buffer or get the silence midle between the previous sentence and the current one
sentence_start = max(sentence_start - buffer, (previous_word_end + sentence_start)/2)
sentence_start = max(sentence_start - buffer, (previous_word_end + sentence_start) / 2)
sentence = word.word
first_word = False
@ -124,13 +128,10 @@ def format_audio_list(audio_files, target_language="en", out_path=None, buffer=0
i += 1
first_word = True
audio = wav[int(sr*sentence_start):int(sr*word_end)].unsqueeze(0)
audio = wav[int(sr * sentence_start) : int(sr * word_end)].unsqueeze(0)
# if the audio is too short ignore it (i.e < 0.33 seconds)
if audio.size(-1) >= sr/3:
torchaudio.save(absoulte_path,
audio,
sr
)
if audio.size(-1) >= sr / 3:
torchaudio.save(absoulte_path, audio, sr)
else:
continue
@ -140,17 +141,17 @@ def format_audio_list(audio_files, target_language="en", out_path=None, buffer=0
df = pandas.DataFrame(metadata)
df = df.sample(frac=1)
num_val_samples = int(len(df)*eval_percentage)
num_val_samples = int(len(df) * eval_percentage)
df_eval = df[:num_val_samples]
df_train = df[num_val_samples:]
df_train = df_train.sort_values('audio_file')
df_train = df_train.sort_values("audio_file")
train_metadata_path = os.path.join(out_path, "metadata_train.csv")
df_train.to_csv(train_metadata_path, sep="|", index=False)
eval_metadata_path = os.path.join(out_path, "metadata_eval.csv")
df_eval = df_eval.sort_values('audio_file')
df_eval = df_eval.sort_values("audio_file")
df_eval.to_csv(eval_metadata_path, sep="|", index=False)
# deallocate VRAM and RAM

22
TTS/demos/xtts_ft_demo/utils/gpt_train.py
View File

@ -1,11 +1,12 @@
import os
import gc
import os
from trainer import Trainer, TrainerArgs
from TTS.config.shared_configs import BaseDatasetConfig
from TTS.tts.datasets import load_tts_samples
from TTS.tts.layers.xtts.trainer.gpt_trainer import GPTArgs, GPTTrainer, GPTTrainerConfig, XttsAudioConfig
from TTS.tts.layers.xtts.trainer.gpt_trainer import GPTArgs, GPTTrainer, GPTTrainerConfig
from TTS.tts.models.xtts import XttsAudioConfig
from TTS.utils.manage import ModelManager
@ -25,7 +26,6 @@ def train_gpt(language, num_epochs, batch_size, grad_acumm, train_csv, eval_csv,
BATCH_SIZE = batch_size # set here the batch size
GRAD_ACUMM_STEPS = grad_acumm # set here the grad accumulation steps
# Define here the dataset that you want to use for the fine-tuning on.
config_dataset = BaseDatasetConfig(
formatter="coqui",
@ -43,10 +43,9 @@ def train_gpt(language, num_epochs, batch_size, grad_acumm, train_csv, eval_csv,
CHECKPOINTS_OUT_PATH = os.path.join(OUT_PATH, "XTTS_v2.0_original_model_files/")
os.makedirs(CHECKPOINTS_OUT_PATH, exist_ok=True)
# DVAE files
DVAE_CHECKPOINT_LINK = "https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v2/main/dvae.pth"
MEL_NORM_LINK = "https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v2/main/mel_stats.pth"
DVAE_CHECKPOINT_LINK = "https://huggingface.co/coqui/XTTS-v2/resolve/main/dvae.pth"
MEL_NORM_LINK = "https://huggingface.co/coqui/XTTS-v2/resolve/main/mel_stats.pth"
# Set the path to the downloaded files
DVAE_CHECKPOINT = os.path.join(CHECKPOINTS_OUT_PATH, os.path.basename(DVAE_CHECKPOINT_LINK))
@ -55,13 +54,14 @@ def train_gpt(language, num_epochs, batch_size, grad_acumm, train_csv, eval_csv,
# download DVAE files if needed
if not os.path.isfile(DVAE_CHECKPOINT) or not os.path.isfile(MEL_NORM_FILE):
print(" > Downloading DVAE files!")
ModelManager._download_model_files([MEL_NORM_LINK, DVAE_CHECKPOINT_LINK], CHECKPOINTS_OUT_PATH, progress_bar=True)
ModelManager._download_model_files(
[MEL_NORM_LINK, DVAE_CHECKPOINT_LINK], CHECKPOINTS_OUT_PATH, progress_bar=True
)
# Download XTTS v2.0 checkpoint if needed
TOKENIZER_FILE_LINK = "https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v2/main/vocab.json"
XTTS_CHECKPOINT_LINK = "https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v2/main/model.pth"
XTTS_CONFIG_LINK = "https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v2/main/config.json"
TOKENIZER_FILE_LINK = "https://huggingface.co/coqui/XTTS-v2/resolve/main/vocab.json"
XTTS_CHECKPOINT_LINK = "https://huggingface.co/coqui/XTTS-v2/resolve/main/model.pth"
XTTS_CONFIG_LINK = "https://huggingface.co/coqui/XTTS-v2/resolve/main/config.json"
# XTTS transfer learning parameters: You we need to provide the paths of XTTS model checkpoint that you want to do the fine tuning.
TOKENIZER_FILE = os.path.join(CHECKPOINTS_OUT_PATH, os.path.basename(TOKENIZER_FILE_LINK)) # vocab.json file

116
TTS/demos/xtts_ft_demo/xtts_demo.py
View File

@ -1,19 +1,16 @@
import argparse
import logging
import os
import sys
import tempfile
import traceback
import gradio as gr
import librosa.display
import numpy as np
import os
import torch
import torchaudio
import traceback
from TTS.demos.xtts_ft_demo.utils.formatter import format_audio_list
from TTS.demos.xtts_ft_demo.utils.gpt_train import train_gpt
from TTS.tts.configs.xtts_config import XttsConfig
from TTS.tts.models.xtts import Xtts
@ -23,7 +20,10 @@ def clear_gpu_cache():
if torch.cuda.is_available():
torch.cuda.empty_cache()
XTTS_MODEL = None
def load_model(xtts_checkpoint, xtts_config, xtts_vocab):
global XTTS_MODEL
clear_gpu_cache()
@ -40,11 +40,17 @@ def load_model(xtts_checkpoint, xtts_config, xtts_vocab):
print("Model Loaded!")
return "Model Loaded!"
def run_tts(lang, tts_text, speaker_audio_file):
if XTTS_MODEL is None or not speaker_audio_file:
return "You need to run the previous step to load the model !!", None, None
gpt_cond_latent, speaker_embedding = XTTS_MODEL.get_conditioning_latents(audio_path=speaker_audio_file, gpt_cond_len=XTTS_MODEL.config.gpt_cond_len, max_ref_length=XTTS_MODEL.config.max_ref_len, sound_norm_refs=XTTS_MODEL.config.sound_norm_refs)
gpt_cond_latent, speaker_embedding = XTTS_MODEL.get_conditioning_latents(
audio_path=speaker_audio_file,
gpt_cond_len=XTTS_MODEL.config.gpt_cond_len,
max_ref_length=XTTS_MODEL.config.max_ref_len,
sound_norm_refs=XTTS_MODEL.config.sound_norm_refs,
)
out = XTTS_MODEL.inference(
text=tts_text,
language=lang,
@ -65,8 +71,6 @@ def run_tts(lang, tts_text, speaker_audio_file):
return "Speech generated !", out_path, speaker_audio_file
# define a logger to redirect
class Logger:
def __init__(self, filename="log.out"):
@ -85,21 +89,19 @@ class Logger:
def isatty(self):
return False
# redirect stdout and stderr to a file
sys.stdout = Logger()
sys.stderr = sys.stdout
# logging.basicConfig(stream=sys.stdout, level=logging.INFO)
import logging
logging.basicConfig(
level=logging.INFO,
format="%(asctime)s [%(levelname)s] %(message)s",
handlers=[
logging.StreamHandler(sys.stdout)
]
level=logging.INFO, format="%(asctime)s [%(levelname)s] %(message)s", handlers=[logging.StreamHandler(sys.stdout)]
)
def read_logs():
sys.stdout.flush()
with open(sys.stdout.log_file, "r") as f:
@ -107,7 +109,6 @@ def read_logs():
if __name__ == "__main__":
parser = argparse.ArgumentParser(
description="""XTTS fine-tuning demo\n\n"""
"""
@ -190,12 +191,11 @@ if __name__ == "__main__":
"zh",
"hu",
"ko",
"ja"
"ja",
"hi",
],
)
progress_data = gr.Label(
label="Progress:"
)
progress_data = gr.Label(label="Progress:")
logs = gr.Textbox(
label="Logs:",
interactive=False,
@ -209,14 +209,24 @@ if __name__ == "__main__":
out_path = os.path.join(out_path, "dataset")
os.makedirs(out_path, exist_ok=True)
if audio_path is None:
return "You should provide one or multiple audio files! If you provided it, probably the upload of the files is not finished yet!", "", ""
return (
"You should provide one or multiple audio files! If you provided it, probably the upload of the files is not finished yet!",
"",
"",
)
else:
try:
train_meta, eval_meta, audio_total_size = format_audio_list(audio_path, target_language=language, out_path=out_path, gradio_progress=progress)
train_meta, eval_meta, audio_total_size = format_audio_list(
audio_path, target_language=language, out_path=out_path, gradio_progress=progress
)
except:
traceback.print_exc()
error = traceback.format_exc()
return f"The data processing was interrupted due an error !! Please check the console to verify the full error message! \n Error summary: {error}", "", ""
return (
f"The data processing was interrupted due an error !! Please check the console to verify the full error message! \n Error summary: {error}",
"",
"",
)
clear_gpu_cache()
@ -264,9 +274,7 @@ if __name__ == "__main__":
step=1,
value=args.max_audio_length,
)
progress_train = gr.Label(
label="Progress:"
)
progress_train = gr.Label(label="Progress:")
logs_tts_train = gr.Textbox(
label="Logs:",
interactive=False,
@ -274,18 +282,41 @@ if __name__ == "__main__":
demo.load(read_logs, None, logs_tts_train, every=1)
train_btn = gr.Button(value="Step 2 - Run the training")
def train_model(language, train_csv, eval_csv, num_epochs, batch_size, grad_acumm, output_path, max_audio_length):
def train_model(
language, train_csv, eval_csv, num_epochs, batch_size, grad_acumm, output_path, max_audio_length
):
clear_gpu_cache()
if not train_csv or not eval_csv:
return "You need to run the data processing step or manually set `Train CSV` and `Eval CSV` fields !", "", "", "", ""
return (
"You need to run the data processing step or manually set `Train CSV` and `Eval CSV` fields !",
"",
"",
"",
"",
)
try:
# convert seconds to waveform frames
max_audio_length = int(max_audio_length * 22050)
config_path, original_xtts_checkpoint, vocab_file, exp_path, speaker_wav = train_gpt(language, num_epochs, batch_size, grad_acumm, train_csv, eval_csv, output_path=output_path, max_audio_length=max_audio_length)
config_path, original_xtts_checkpoint, vocab_file, exp_path, speaker_wav = train_gpt(
language,
num_epochs,
batch_size,
grad_acumm,
train_csv,
eval_csv,
output_path=output_path,
max_audio_length=max_audio_length,
)
except:
traceback.print_exc()
error = traceback.format_exc()
return f"The training was interrupted due an error !! Please check the console to check the full error message! \n Error summary: {error}", "", "", "", ""
return (
f"The training was interrupted due an error !! Please check the console to check the full error message! \n Error summary: {error}",
"",
"",
"",
"",
)
# copy original files to avoid parameters changes issues
os.system(f"cp {config_path} {exp_path}")
@ -312,9 +343,7 @@ if __name__ == "__main__":
label="XTTS vocab path:",
value="",
)
progress_load = gr.Label(
label="Progress:"
)
progress_load = gr.Label(label="Progress:")
load_btn = gr.Button(value="Step 3 - Load Fine-tuned XTTS model")
with gr.Column() as col2:
@ -342,7 +371,8 @@ if __name__ == "__main__":
"hu",
"ko",
"ja",
]
"hi",
],
)
tts_text = gr.Textbox(
label="Input Text.",
@ -351,9 +381,7 @@ if __name__ == "__main__":
tts_btn = gr.Button(value="Step 4 - Inference")
with gr.Column() as col3:
progress_gen = gr.Label(
label="Progress:"
)
progress_gen = gr.Label(label="Progress:")
tts_output_audio = gr.Audio(label="Generated Audio.")
reference_audio = gr.Audio(label="Reference audio used.")
@ -371,7 +399,6 @@ if __name__ == "__main__":
],
)
train_btn.click(
fn=train_model,
inputs=[
@ -389,11 +416,7 @@ if __name__ == "__main__":
load_btn.click(
fn=load_model,
inputs=[
xtts_checkpoint,
xtts_config,
xtts_vocab
],
inputs=[xtts_checkpoint, xtts_config, xtts_vocab],
outputs=[progress_load],
)
@ -407,9 +430,4 @@ if __name__ == "__main__":
outputs=[progress_gen, tts_output_audio, reference_audio],
)
demo.launch(
share=True,
debug=False,
server_port=args.port,
server_name="0.0.0.0"
)
demo.launch(share=True, debug=False, server_port=args.port, server_name="0.0.0.0")

2
TTS/encoder/README.md
View File

@ -14,5 +14,5 @@ To run the code, you need to follow the same flow as in TTS.
- Define 'config.json' for your needs. Note that, audio parameters should match your TTS model.
- Example training call ```python speaker_encoder/train.py --config_path speaker_encoder/config.json --data_path ~/Data/Libri-TTS/train-clean-360```
- Generate embedding vectors ```python speaker_encoder/compute_embeddings.py --use_cuda true /model/path/best_model.pth model/config/path/config.json dataset/path/ output_path``` . This code parses all .wav files at the given dataset path and generates the same folder structure under the output path with the generated embedding files.
- Generate embedding vectors ```python speaker_encoder/compute_embeddings.py --use_cuda /model/path/best_model.pth model/config/path/config.json dataset/path/ output_path``` . This code parses all .wav files at the given dataset path and generates the same folder structure under the output path with the generated embedding files.
- Watch training on Tensorboard as in TTS

2
TTS/encoder/configs/emotion_encoder_config.py
View File

@ -1,4 +1,4 @@
from dataclasses import asdict, dataclass
from dataclasses import dataclass
from TTS.encoder.configs.base_encoder_config import BaseEncoderConfig

2
TTS/encoder/configs/speaker_encoder_config.py
View File

@ -1,4 +1,4 @@
from dataclasses import asdict, dataclass
from dataclasses import dataclass
from TTS.encoder.configs.base_encoder_config import BaseEncoderConfig

19
TTS/encoder/dataset.py
View File

@ -1,3 +1,4 @@
import logging
import random
import torch
@ -5,6 +6,8 @@ from torch.utils.data import Dataset
from TTS.encoder.utils.generic_utils import AugmentWAV
logger = logging.getLogger(__name__)
class EncoderDataset(Dataset):
def __init__(
@ -15,7 +18,6 @@ class EncoderDataset(Dataset):
voice_len=1.6,
num_classes_in_batch=64,
num_utter_per_class=10,
verbose=False,
augmentation_config=None,
use_torch_spec=None,
):
@ -24,7 +26,6 @@ class EncoderDataset(Dataset):
ap (TTS.tts.utils.AudioProcessor): audio processor object.
meta_data (list): list of dataset instances.
seq_len (int): voice segment length in seconds.
verbose (bool): print diagnostic information.
"""
super().__init__()
self.config = config
@ -33,7 +34,6 @@ class EncoderDataset(Dataset):
self.seq_len = int(voice_len * self.sample_rate)
self.num_utter_per_class = num_utter_per_class
self.ap = ap
self.verbose = verbose
self.use_torch_spec = use_torch_spec
self.classes, self.items = self.__parse_items()
@ -50,13 +50,12 @@ class EncoderDataset(Dataset):
if "gaussian" in augmentation_config.keys():
self.gaussian_augmentation_config = augmentation_config["gaussian"]
if self.verbose:
print("\n > DataLoader initialization")
print(f" | > Classes per Batch: {num_classes_in_batch}")
print(f" | > Number of instances : {len(self.items)}")
print(f" | > Sequence length: {self.seq_len}")
print(f" | > Num Classes: {len(self.classes)}")
print(f" | > Classes: {self.classes}")
logger.info("DataLoader initialization")
logger.info(" | Classes per batch: %d", num_classes_in_batch)
logger.info(" | Number of instances: %d", len(self.items))
logger.info(" | Sequence length: %d", self.seq_len)
logger.info(" | Number of classes: %d", len(self.classes))
logger.info(" | Classes: %s", self.classes)
def load_wav(self, filename):
audio = self.ap.load_wav(filename, sr=self.ap.sample_rate)

12
TTS/encoder/losses.py
View File

@ -1,7 +1,11 @@
import logging
import torch
import torch.nn.functional as F
from torch import nn
logger = logging.getLogger(__name__)
# adapted from https://github.com/cvqluu/GE2E-Loss
class GE2ELoss(nn.Module):
@ -23,7 +27,7 @@ class GE2ELoss(nn.Module):
self.b = nn.Parameter(torch.tensor(init_b))
self.loss_method = loss_method
print(" > Initialized Generalized End-to-End loss")
logger.info("Initialized Generalized End-to-End loss")
assert self.loss_method in ["softmax", "contrast"]
@ -139,7 +143,7 @@ class AngleProtoLoss(nn.Module):
self.b = nn.Parameter(torch.tensor(init_b))
self.criterion = torch.nn.CrossEntropyLoss()
print(" > Initialized Angular Prototypical loss")
logger.info("Initialized Angular Prototypical loss")
def forward(self, x, _label=None):
"""
@ -177,7 +181,7 @@ class SoftmaxLoss(nn.Module):
self.criterion = torch.nn.CrossEntropyLoss()
self.fc = nn.Linear(embedding_dim, n_speakers)
print("Initialised Softmax Loss")
logger.info("Initialised Softmax Loss")
def forward(self, x, label=None):
# reshape for compatibility
@ -212,7 +216,7 @@ class SoftmaxAngleProtoLoss(nn.Module):
self.softmax = SoftmaxLoss(embedding_dim, n_speakers)
self.angleproto = AngleProtoLoss(init_w, init_b)
print("Initialised SoftmaxAnglePrototypical Loss")
logger.info("Initialised SoftmaxAnglePrototypical Loss")
def forward(self, x, label=None):
"""

16
TTS/encoder/models/base_encoder.py
View File

@ -1,12 +1,16 @@
import logging
import numpy as np
import torch
import torchaudio
from coqpit import Coqpit
from torch import nn
from trainer.generic_utils import set_partial_state_dict
from trainer.io import load_fsspec
from TTS.encoder.losses import AngleProtoLoss, GE2ELoss, SoftmaxAngleProtoLoss
from TTS.utils.generic_utils import set_init_dict
from TTS.utils.io import load_fsspec
logger = logging.getLogger(__name__)
class PreEmphasis(nn.Module):
@ -118,15 +122,15 @@ class BaseEncoder(nn.Module):
state = load_fsspec(checkpoint_path, map_location=torch.device("cpu"), cache=cache)
try:
self.load_state_dict(state["model"])
print(" > Model fully restored. ")
logger.info("Model fully restored. ")
except (KeyError, RuntimeError) as error:
# If eval raise the error
if eval:
raise error
print(" > Partial model initialization.")
logger.info("Partial model initialization.")
model_dict = self.state_dict()
model_dict = set_init_dict(model_dict, state["model"], c)
model_dict = set_partial_state_dict(model_dict, state["model"], config)
self.load_state_dict(model_dict)
del model_dict
@ -135,7 +139,7 @@ class BaseEncoder(nn.Module):
try:
criterion.load_state_dict(state["criterion"])
except (KeyError, RuntimeError) as error:
print(" > Criterion load ignored because of:", error)
logger.exception("Criterion load ignored because of: %s", error)
# instance and load the criterion for the encoder classifier in inference time
if (

2
TTS/encoder/models/lstm.py
View File

@ -86,7 +86,7 @@ class LSTMSpeakerEncoder(BaseEncoder):
- x: :math:`(N, 1, T_{in})` or :math:`(N, D_{spec}, T_{in})`
"""
with torch.no_grad():
with torch.cuda.amp.autocast(enabled=False):
with torch.autocast("cuda", enabled=False):
if self.use_torch_spec:
x.squeeze_(1)
x = self.torch_spec(x)

13
TTS/encoder/utils/generic_utils.py
View File

@ -1,4 +1,5 @@
import glob
import logging
import os
import random
@ -8,6 +9,8 @@ from scipy import signal
from TTS.encoder.models.lstm import LSTMSpeakerEncoder
from TTS.encoder.models.resnet import ResNetSpeakerEncoder
logger = logging.getLogger(__name__)
class AugmentWAV(object):
def __init__(self, ap, augmentation_config):
@ -34,12 +37,14 @@ class AugmentWAV(object):
# ignore not listed directories
if noise_dir not in self.additive_noise_types:
continue
if not noise_dir in self.noise_list:
if noise_dir not in self.noise_list:
self.noise_list[noise_dir] = []
self.noise_list[noise_dir].append(wav_file)
print(
f" | > Using Additive Noise Augmentation: with {len(additive_files)} audios instances from {self.additive_noise_types}"
logger.info(
"Using Additive Noise Augmentation: with %d audios instances from %s",
len(additive_files),
self.additive_noise_types,
)
self.use_rir = False
@ -50,7 +55,7 @@ class AugmentWAV(object):
self.rir_files = glob.glob(os.path.join(self.rir_config["rir_path"], "**/*.wav"), recursive=True)
self.use_rir = True
print(f" | > Using RIR Noise Augmentation: with {len(self.rir_files)} audios instances")
logger.info("Using RIR Noise Augmentation: with %d audios instances", len(self.rir_files))
self.create_augmentation_global_list()

41
TTS/encoder/utils/prepare_voxceleb.py
View File

@ -19,15 +19,19 @@
# pylint: disable=too-many-locals, too-many-statements, too-many-arguments, too-many-instance-attributes
""" voxceleb 1 & 2 """
import csv
import hashlib
import logging
import os
import subprocess
import sys
import zipfile
import pandas
import soundfile as sf
from absl import logging
from TTS.utils.generic_utils import ConsoleFormatter, setup_logger
logger = logging.getLogger(__name__)
SUBSETS = {
"vox1_dev_wav": [
@ -77,14 +81,14 @@ def download_and_extract(directory, subset, urls):
zip_filepath = os.path.join(directory, url.split("/")[-1])
if os.path.exists(zip_filepath):
continue
logging.info("Downloading %s to %s" % (url, zip_filepath))
logger.info("Downloading %s to %s" % (url, zip_filepath))
subprocess.call(
"wget %s --user %s --password %s -O %s" % (url, USER["user"], USER["password"], zip_filepath),
shell=True,
)
statinfo = os.stat(zip_filepath)
logging.info("Successfully downloaded %s, size(bytes): %d" % (url, statinfo.st_size))
logger.info("Successfully downloaded %s, size(bytes): %d" % (url, statinfo.st_size))
# concatenate all parts into zip files
if ".zip" not in zip_filepath:
@ -118,9 +122,9 @@ def exec_cmd(cmd):
try:
retcode = subprocess.call(cmd, shell=True)
if retcode < 0:
logging.info(f"Child was terminated by signal {retcode}")
logger.info(f"Child was terminated by signal {retcode}")
except OSError as e:
logging.info(f"Execution failed: {e}")
logger.info(f"Execution failed: {e}")
retcode = -999
return retcode
@ -134,11 +138,11 @@ def decode_aac_with_ffmpeg(aac_file, wav_file):
bool, True if success.
"""
cmd = f"ffmpeg -i {aac_file} {wav_file}"
logging.info(f"Decoding aac file using command line: {cmd}")
logger.info(f"Decoding aac file using command line: {cmd}")
ret = exec_cmd(cmd)
if ret != 0:
logging.error(f"Failed to decode aac file with retcode {ret}")
logging.error("Please check your ffmpeg installation.")
logger.error(f"Failed to decode aac file with retcode {ret}")
logger.error("Please check your ffmpeg installation.")
return False
return True
@ -152,7 +156,7 @@ def convert_audio_and_make_label(input_dir, subset, output_dir, output_file):
output_file: the name of the newly generated csv file. e.g. vox1_dev_wav.csv
"""
logging.info("Preprocessing audio and label for subset %s" % subset)
logger.info("Preprocessing audio and label for subset %s" % subset)
source_dir = os.path.join(input_dir, subset)
files = []
@ -185,9 +189,12 @@ def convert_audio_and_make_label(input_dir, subset, output_dir, output_file):
# Write to CSV file which contains four columns:
# "wav_filename", "wav_length_ms", "speaker_id", "speaker_name".
csv_file_path = os.path.join(output_dir, output_file)
df = pandas.DataFrame(data=files, columns=["wav_filename", "wav_length_ms", "speaker_id", "speaker_name"])
df.to_csv(csv_file_path, index=False, sep="\t")
logging.info("Successfully generated csv file {}".format(csv_file_path))
with open(csv_file_path, "w", newline="", encoding="utf-8") as f:
writer = csv.writer(f, delimiter="\t")
writer.writerow(["wav_filename", "wav_length_ms", "speaker_id", "speaker_name"])
for wav_file in files:
writer.writerow(wav_file)
logger.info("Successfully generated csv file {}".format(csv_file_path))
def processor(directory, subset, force_process):
@ -200,16 +207,16 @@ def processor(directory, subset, force_process):
if not force_process and os.path.exists(subset_csv):
return subset_csv
logging.info("Downloading and process the voxceleb in %s", directory)
logging.info("Preparing subset %s", subset)
logger.info("Downloading and process the voxceleb in %s", directory)
logger.info("Preparing subset %s", subset)
download_and_extract(directory, subset, urls[subset])
convert_audio_and_make_label(directory, subset, directory, subset + ".csv")
logging.info("Finished downloading and processing")
logger.info("Finished downloading and processing")
return subset_csv
if __name__ == "__main__":
logging.set_verbosity(logging.INFO)
setup_logger("TTS", level=logging.INFO, stream=sys.stdout, formatter=ConsoleFormatter())
if len(sys.argv) != 4:
print("Usage: python prepare_data.py save_directory user password")
sys.exit()

8
TTS/encoder/utils/training.py
View File

@ -2,14 +2,14 @@ import os
from dataclasses import dataclass, field
from coqpit import Coqpit
from trainer import TrainerArgs, get_last_checkpoint
from trainer.io import copy_model_files
from trainer import TrainerArgs
from trainer.generic_utils import get_experiment_folder_path, get_git_branch
from trainer.io import copy_model_files, get_last_checkpoint
from trainer.logging import logger_factory
from trainer.logging.console_logger import ConsoleLogger
from TTS.config import load_config, register_config
from TTS.tts.utils.text.characters import parse_symbols
from TTS.utils.generic_utils import get_experiment_folder_path, get_git_branch
@dataclass
@ -29,7 +29,7 @@ def process_args(args, config=None):
args (argparse.Namespace or dict like): Parsed input arguments.
config (Coqpit): Model config. If none, it is generated from `args`. Defaults to None.
Returns:
c (TTS.utils.io.AttrDict): Config paramaters.
c (Coqpit): Config paramaters.
out_path (str): Path to save models and logging.
audio_path (str): Path to save generated test audios.
c_logger (TTS.utils.console_logger.ConsoleLogger): Class that does

5
TTS/encoder/utils/visual.py
View File

@ -1,7 +1,6 @@
import matplotlib
import matplotlib.pyplot as plt
import numpy as np
import umap
matplotlib.use("Agg")
@ -30,6 +29,10 @@ colormap = (
def plot_embeddings(embeddings, num_classes_in_batch):
try:
import umap
except ImportError as e:
raise ImportError("Package not installed: umap-learn") from e
num_utter_per_class = embeddings.shape[0] // num_classes_in_batch
# if necessary get just the first 10 classes

23
TTS/model.py
View File

@ -1,5 +1,6 @@
import os
from abc import abstractmethod
from typing import Dict
from typing import Any, Union
import torch
from coqpit import Coqpit
@ -11,12 +12,12 @@ from trainer import TrainerModel
class BaseTrainerModel(TrainerModel):
"""BaseTrainerModel model expanding TrainerModel with required functions by 🐸TTS.
Every new 🐸TTS model must inherit it.
Every new Coqui model must inherit it.
"""
@staticmethod
@abstractmethod
def init_from_config(config: Coqpit):
def init_from_config(config: Coqpit) -> "BaseTrainerModel":
"""Init the model and all its attributes from the given config.
Override this depending on your model.
@ -24,7 +25,7 @@ class BaseTrainerModel(TrainerModel):
...
@abstractmethod
def inference(self, input: torch.Tensor, aux_input={}) -> Dict:
def inference(self, input: torch.Tensor, aux_input: dict[str, Any] = {}) -> dict[str, Any]:
"""Forward pass for inference.
It must return a dictionary with the main model output and all the auxiliary outputs. The key ```model_outputs```
@ -45,15 +46,21 @@ class BaseTrainerModel(TrainerModel):
@abstractmethod
def load_checkpoint(
self, config: Coqpit, checkpoint_path: str, eval: bool = False, strict: bool = True, cache=False
self,
config: Coqpit,
checkpoint_path: Union[str, os.PathLike[Any]],
eval: bool = False,
strict: bool = True,
cache: bool = False,
) -> None:
"""Load a model checkpoint gile and get ready for training or inference.
"""Load a model checkpoint file and get ready for training or inference.
Args:
config (Coqpit): Model configuration.
checkpoint_path (str): Path to the model checkpoint file.
checkpoint_path (str | os.PathLike): Path to the model checkpoint file.
eval (bool, optional): If true, init model for inference else for training. Defaults to False.
strict (bool, optional): Match all checkpoint keys to model's keys. Defaults to True.
cache (bool, optional): If True, cache the file locally for subsequent calls. It is cached under `get_user_data_dir()/tts_cache`. Defaults to False.
cache (bool, optional): If True, cache the file locally for subsequent calls.
It is cached under `trainer.io.get_user_data_dir()/tts_cache`. Defaults to False.
"""
...

7
TTS/server/README.md
View File

@ -1,5 +1,8 @@
# :frog: TTS demo server
Before you use the server, make sure you [install](https://github.com/coqui-ai/TTS/tree/dev#install-tts)) :frog: TTS properly. Then, you can follow the steps below.
Before you use the server, make sure you
[install](https://github.com/idiap/coqui-ai-TTS/tree/dev#install-tts)) :frog: TTS
properly and install the additional dependencies with `pip install
coqui-tts[server]`. Then, you can follow the steps below.
**Note:** If you install :frog:TTS using ```pip```, you can also use the ```tts-server``` end point on the terminal.
@ -12,7 +15,7 @@ Run the server with the official models.
```python TTS/server/server.py --model_name tts_models/en/ljspeech/tacotron2-DCA --vocoder_name vocoder_models/en/ljspeech/multiband-melgan```
Run the server with the official models on a GPU.
```CUDA_VISIBLE_DEVICES="0" python TTS/server/server.py --model_name tts_models/en/ljspeech/tacotron2-DCA --vocoder_name vocoder_models/en/ljspeech/multiband-melgan --use_cuda True```
```CUDA_VISIBLE_DEVICES="0" python TTS/server/server.py --model_name tts_models/en/ljspeech/tacotron2-DCA --vocoder_name vocoder_models/en/ljspeech/multiband-melgan --use_cuda```
Run the server with a custom models.
```python TTS/server/server.py --tts_checkpoint /path/to/tts/model.pth --tts_config /path/to/tts/config.json --vocoder_checkpoint /path/to/vocoder/model.pth --vocoder_config /path/to/vocoder/config.json```

50
TTS/server/server.py
View File

@ -1,7 +1,11 @@
#!flask/bin/python
"""TTS demo server."""
import argparse
import io
import json
import logging
import os
import sys
from pathlib import Path
@ -9,24 +13,26 @@ from threading import Lock
from typing import Union
from urllib.parse import parse_qs
from flask import Flask, render_template, render_template_string, request, send_file
try:
from flask import Flask, render_template, render_template_string, request, send_file
except ImportError as e:
msg = "Server requires requires flask, use `pip install coqui-tts[server]`"
raise ImportError(msg) from e
from TTS.config import load_config
from TTS.utils.generic_utils import ConsoleFormatter, setup_logger
from TTS.utils.manage import ModelManager
from TTS.utils.synthesizer import Synthesizer
logger = logging.getLogger(__name__)
setup_logger("TTS", level=logging.INFO, stream=sys.stdout, formatter=ConsoleFormatter())
def create_argparser():
def convert_boolean(x):
return x.lower() in ["true", "1", "yes"]
def create_argparser() -> argparse.ArgumentParser:
parser = argparse.ArgumentParser()
parser.add_argument(
"--list_models",
type=convert_boolean,
nargs="?",
const=True,
default=False,
action="store_true",
help="list available pre-trained tts and vocoder models.",
)
parser.add_argument(
@ -54,9 +60,13 @@ def create_argparser():
parser.add_argument("--vocoder_config_path", type=str, help="Path to vocoder model config file.", default=None)
parser.add_argument("--speakers_file_path", type=str, help="JSON file for multi-speaker model.", default=None)
parser.add_argument("--port", type=int, default=5002, help="port to listen on.")
parser.add_argument("--use_cuda", type=convert_boolean, default=False, help="true to use CUDA.")
parser.add_argument("--debug", type=convert_boolean, default=False, help="true to enable Flask debug mode.")
parser.add_argument("--show_details", type=convert_boolean, default=False, help="Generate model detail page.")
parser.add_argument("--use_cuda", action=argparse.BooleanOptionalAction, default=False, help="true to use CUDA.")
parser.add_argument(
"--debug", action=argparse.BooleanOptionalAction, default=False, help="true to enable Flask debug mode."
)
parser.add_argument(
"--show_details", action=argparse.BooleanOptionalAction, default=False, help="Generate model detail page."
)
return parser
@ -66,10 +76,6 @@ args = create_argparser().parse_args()
path = Path(__file__).parent / "../.models.json"
manager = ModelManager(path)
if args.list_models:
manager.list_models()
sys.exit()
# update in-use models to the specified released models.
model_path = None
config_path = None
@ -164,14 +170,12 @@ def index():
def details():
if args.config_path is not None and os.path.isfile(args.config_path):
model_config = load_config(args.config_path)
else:
if args.model_name is not None:
elif args.model_name is not None:
model_config = load_config(config_path)
if args.vocoder_config_path is not None and os.path.isfile(args.vocoder_config_path):
vocoder_config = load_config(args.vocoder_config_path)
else:
if args.vocoder_name is not None:
elif args.vocoder_name is not None:
vocoder_config = load_config(vocoder_config_path)
else:
vocoder_config = None
@ -197,9 +201,9 @@ def tts():
style_wav = request.headers.get("style-wav") or request.values.get("style_wav", "")
style_wav = style_wav_uri_to_dict(style_wav)
print(f" > Model input: {text}")
print(f" > Speaker Idx: {speaker_idx}")
print(f" > Language Idx: {language_idx}")
logger.info("Model input: %s", text)
logger.info("Speaker idx: %s", speaker_idx)
logger.info("Language idx: %s", language_idx)
wavs = synthesizer.tts(text, speaker_name=speaker_idx, language_name=language_idx, style_wav=style_wav)
out = io.BytesIO()
synthesizer.save_wav(wavs, out)
@ -243,7 +247,7 @@ def mary_tts_api_process():
text = data.get("INPUT_TEXT", [""])[0]
else:
text = request.args.get("INPUT_TEXT", "")
print(f" > Model input: {text}")
logger.info("Model input: %s", text)
wavs = synthesizer.tts(text)
out = io.BytesIO()
synthesizer.save_wav(wavs, out)

2
TTS/server/templates/index.html
View File

@ -30,7 +30,7 @@
</head>
<body>
<a href="https://github.com/coqui-ai/TTS"><img style="position: absolute; z-index:1000; top: 0; left: 0; border: 0;"
<a href="https://github.com/idiap/coqui-ai-TTS"><img style="position: absolute; z-index:1000; top: 0; left: 0; border: 0;"
src="https://s3.amazonaws.com/github/ribbons/forkme_left_darkblue_121621.png" alt="Fork me on GitHub"></a>
<!-- Navigation -->

4
TTS/tts/configs/bark_config.py
View File

@ -2,11 +2,12 @@ import os
from dataclasses import dataclass, field
from typing import Dict
from trainer.io import get_user_data_dir
from TTS.tts.configs.shared_configs import BaseTTSConfig
from TTS.tts.layers.bark.model import GPTConfig
from TTS.tts.layers.bark.model_fine import FineGPTConfig
from TTS.tts.models.bark import BarkAudioConfig
from TTS.utils.generic_utils import get_user_data_dir
@dataclass
@ -95,7 +96,6 @@ class BarkConfig(BaseTTSConfig):
"coarse": os.path.join(self.CACHE_DIR, "coarse_2.pt"),
"fine": os.path.join(self.CACHE_DIR, "fine_2.pt"),
"hubert_tokenizer": os.path.join(self.CACHE_DIR, "tokenizer.pth"),
"hubert": os.path.join(self.CACHE_DIR, "hubert.pt"),
}
self.SMALL_REMOTE_MODEL_PATHS = {
"text": {"path": os.path.join(self.REMOTE_BASE_URL, "text.pt")},

23
TTS/tts/datasets/__init__.py
View File

@ -1,3 +1,4 @@
import logging
import os
import sys
from collections import Counter
@ -9,6 +10,8 @@ import numpy as np
from TTS.tts.datasets.dataset import *
from TTS.tts.datasets.formatters import *
logger = logging.getLogger(__name__)
def split_dataset(items, eval_split_max_size=None, eval_split_size=0.01):
"""Split a dataset into train and eval. Consider speaker distribution in multi-speaker training.
@ -122,7 +125,7 @@ def load_tts_samples(
meta_data_train = add_extra_keys(meta_data_train, language, dataset_name)
print(f" | > Found {len(meta_data_train)} files in {Path(root_path).resolve()}")
logger.info("Found %d files in %s", len(meta_data_train), Path(root_path).resolve())
# load evaluation split if set
if eval_split:
if meta_file_val:
@ -163,19 +166,23 @@ def load_attention_mask_meta_data(metafile_path):
def _get_formatter_by_name(name):
"""Returns the respective preprocessing function."""
thismodule = sys.modules[__name__]
if not hasattr(thismodule, name.lower()):
msg = (
f"{name} formatter not found. If it is a custom formatter, pass the function to load_tts_samples() instead."
)
raise ValueError(msg)
return getattr(thismodule, name.lower())
def find_unique_chars(data_samples, verbose=True):
texts = "".join(item[0] for item in data_samples)
def find_unique_chars(data_samples):
texts = "".join(item["text"] for item in data_samples)
chars = set(texts)
lower_chars = filter(lambda c: c.islower(), chars)
chars_force_lower = [c.lower() for c in chars]
chars_force_lower = set(chars_force_lower)
if verbose:
print(f" > Number of unique characters: {len(chars)}")
print(f" > Unique characters: {''.join(sorted(chars))}")
print(f" > Unique lower characters: {''.join(sorted(lower_chars))}")
print(f" > Unique all forced to lower characters: {''.join(sorted(chars_force_lower))}")
logger.info("Number of unique characters: %d", len(chars))
logger.info("Unique characters: %s", "".join(sorted(chars)))
logger.info("Unique lower characters: %s", "".join(sorted(lower_chars)))
logger.info("Unique all forced to lower characters: %s", "".join(sorted(chars_force_lower)))
return chars_force_lower

289
TTS/tts/datasets/dataset.py
View File

@ -1,11 +1,14 @@
import base64
import collections
import logging
import os
import random
from typing import Dict, List, Union
from typing import Any, Optional, Union
import numpy as np
import numpy.typing as npt
import torch
import torchaudio
import tqdm
from torch.utils.data import Dataset
@ -13,7 +16,7 @@ from TTS.tts.utils.data import prepare_data, prepare_stop_target, prepare_tensor
from TTS.utils.audio import AudioProcessor
from TTS.utils.audio.numpy_transforms import compute_energy as calculate_energy
import mutagen
logger = logging.getLogger(__name__)
# to prevent too many open files error as suggested here
# https://github.com/pytorch/pytorch/issues/11201#issuecomment-421146936
@ -30,27 +33,59 @@ def _parse_sample(item):
elif len(item) == 3:
text, wav_file, speaker_name = item
else:
raise ValueError(" [!] Dataset cannot parse the sample.")
msg = "Dataset cannot parse the sample."
raise ValueError(msg)
return text, wav_file, speaker_name, language_name, attn_file
def noise_augment_audio(wav):
def noise_augment_audio(wav: npt.NDArray) -> npt.NDArray:
return wav + (1.0 / 32768.0) * np.random.rand(*wav.shape)
def string2filename(string):
def string2filename(string: str) -> str:
# generate a safe and reversible filename based on a string
filename = base64.urlsafe_b64encode(string.encode("utf-8")).decode("utf-8", "ignore")
return filename
return base64.urlsafe_b64encode(string.encode("utf-8")).decode("utf-8", "ignore")
def get_audio_size(audiopath):
def get_audio_size(audiopath: Union[str, os.PathLike[Any]]) -> int:
"""Return the number of samples in the audio file."""
if not isinstance(audiopath, str):
audiopath = str(audiopath)
extension = audiopath.rpartition(".")[-1].lower()
if extension not in {"mp3", "wav", "flac"}:
raise RuntimeError(f"The audio format {extension} is not supported, please convert the audio files to mp3, flac, or wav format!")
msg = f"The audio format {extension} is not supported, please convert the audio files to mp3, flac, or wav format!"
raise RuntimeError(msg)
audio_info = mutagen.File(audiopath).info
return int(audio_info.length * audio_info.sample_rate)
try:
return torchaudio.info(audiopath).num_frames
except RuntimeError as e:
msg = f"Failed to decode {audiopath}"
raise RuntimeError(msg) from e
def get_attribute_balancer_weights(items: list, attr_name: str, multi_dict: Optional[dict] = None):
"""Create inverse frequency weights for balancing the dataset.
Use `multi_dict` to scale relative weights."""
attr_names_samples = np.array([item[attr_name] for item in items])
unique_attr_names = np.unique(attr_names_samples).tolist()
attr_idx = [unique_attr_names.index(l) for l in attr_names_samples]
attr_count = np.array([len(np.where(attr_names_samples == l)[0]) for l in unique_attr_names])
weight_attr = 1.0 / attr_count
dataset_samples_weight = np.array([weight_attr[l] for l in attr_idx])
dataset_samples_weight = dataset_samples_weight / np.linalg.norm(dataset_samples_weight)
if multi_dict is not None:
# check if all keys are in the multi_dict
for k in multi_dict:
assert k in unique_attr_names, f"{k} not in {unique_attr_names}"
# scale weights
multiplier_samples = np.array([multi_dict.get(item[attr_name], 1.0) for item in items])
dataset_samples_weight *= multiplier_samples
return (
torch.from_numpy(dataset_samples_weight).float(),
unique_attr_names,
np.unique(dataset_samples_weight).tolist(),
)
class TTSDataset(Dataset):
@ -59,32 +94,32 @@ class TTSDataset(Dataset):
outputs_per_step: int = 1,
compute_linear_spec: bool = False,
ap: AudioProcessor = None,
samples: List[Dict] = None,
samples: Optional[list[dict]] = None,
tokenizer: "TTSTokenizer" = None,
compute_f0: bool = False,
compute_energy: bool = False,
f0_cache_path: str = None,
energy_cache_path: str = None,
f0_cache_path: Optional[str] = None,
energy_cache_path: Optional[str] = None,
return_wav: bool = False,
batch_group_size: int = 0,
min_text_len: int = 0,
max_text_len: int = float("inf"),
min_audio_len: int = 0,
max_audio_len: int = float("inf"),
phoneme_cache_path: str = None,
phoneme_cache_path: Optional[str] = None,
precompute_num_workers: int = 0,
speaker_id_mapping: Dict = None,
d_vector_mapping: Dict = None,
language_id_mapping: Dict = None,
speaker_id_mapping: Optional[dict] = None,
d_vector_mapping: Optional[dict] = None,
language_id_mapping: Optional[dict] = None,
use_noise_augment: bool = False,
start_by_longest: bool = False,
verbose: bool = False,
):
) -> None:
"""Generic 📂 data loader for `tts` models. It is configurable for different outputs and needs.
If you need something different, you can subclass and override.
Args:
----
outputs_per_step (int): Number of time frames predicted per step.
compute_linear_spec (bool): compute linear spectrogram if True.
@ -137,7 +172,6 @@ class TTSDataset(Dataset):
start_by_longest (bool): Start by longest sequence. It is especially useful to check OOM. Defaults to False.
verbose (bool): Print diagnostic information. Defaults to false.
"""
super().__init__()
self.batch_group_size = batch_group_size
@ -161,33 +195,44 @@ class TTSDataset(Dataset):
self.use_noise_augment = use_noise_augment
self.start_by_longest = start_by_longest
self.verbose = verbose
self.rescue_item_idx = 1
self.pitch_computed = False
self.tokenizer = tokenizer
if self.tokenizer.use_phonemes:
self.phoneme_dataset = PhonemeDataset(
self.samples, self.tokenizer, phoneme_cache_path, precompute_num_workers=precompute_num_workers
self.samples,
self.tokenizer,
phoneme_cache_path,
precompute_num_workers=precompute_num_workers,
)
if compute_f0:
self.f0_dataset = F0Dataset(
self.samples, self.ap, cache_path=f0_cache_path, precompute_num_workers=precompute_num_workers
self.samples,
self.ap,
cache_path=f0_cache_path,
precompute_num_workers=precompute_num_workers,
)
if compute_energy:
self.energy_dataset = EnergyDataset(
self.samples, self.ap, cache_path=energy_cache_path, precompute_num_workers=precompute_num_workers
self.samples,
self.ap,
cache_path=energy_cache_path,
precompute_num_workers=precompute_num_workers,
)
if self.verbose:
self.print_logs()
@property
def lengths(self):
def lengths(self) -> list[int]:
lens = []
for item in self.samples:
_, wav_file, *_ = _parse_sample(item)
try:
audio_len = get_audio_size(wav_file)
except RuntimeError:
logger.warning(f"Failed to compute length for {item['audio_file']}")
audio_len = 0
lens.append(audio_len)
return lens
@ -196,7 +241,7 @@ class TTSDataset(Dataset):
return self._samples
@samples.setter
def samples(self, new_samples):
def samples(self, new_samples) -> None:
self._samples = new_samples
if hasattr(self, "f0_dataset"):
self.f0_dataset.samples = new_samples
@ -205,7 +250,7 @@ class TTSDataset(Dataset):
if hasattr(self, "phoneme_dataset"):
self.phoneme_dataset.samples = new_samples
def __len__(self):
def __len__(self) -> int:
return len(self.samples)
def __getitem__(self, idx):
@ -213,11 +258,10 @@ class TTSDataset(Dataset):
def print_logs(self, level: int = 0) -> None:
indent = "\t" * level
print("\n")
print(f"{indent}> DataLoader initialization")
print(f"{indent}| > Tokenizer:")
logger.info("%sDataLoader initialization", indent)
logger.info("%s| Tokenizer:", indent)
self.tokenizer.print_logs(level + 1)
print(f"{indent}| > Number of instances : {len(self.samples)}")
logger.info("%s| Number of instances : %d", indent, len(self.samples))
def load_wav(self, filename):
waveform = self.ap.load_wav(filename)
@ -253,7 +297,7 @@ class TTSDataset(Dataset):
token_ids = self.tokenizer.text_to_ids(text)
return np.array(token_ids, dtype=np.int32)
def load_data(self, idx):
def load_data(self, idx) -> dict[str, Any]:
item = self.samples[idx]
raw_text = item["text"]
@ -287,7 +331,7 @@ class TTSDataset(Dataset):
if self.compute_energy:
energy = self.get_energy(idx)["energy"]
sample = {
return {
"raw_text": raw_text,
"token_ids": token_ids,
"wav": wav,
@ -300,13 +344,16 @@ class TTSDataset(Dataset):
"wav_file_name": os.path.basename(item["audio_file"]),
"audio_unique_name": item["audio_unique_name"],
}
return sample
@staticmethod
def _compute_lengths(samples):
new_samples = []
for item in samples:
try:
audio_length = get_audio_size(item["audio_file"])
except RuntimeError:
logger.warning(f"Failed to compute length, skipping {item['audio_file']}")
continue
text_lenght = len(item["text"])
item["audio_length"] = audio_length
item["text_length"] = text_lenght
@ -314,7 +361,7 @@ class TTSDataset(Dataset):
return new_samples
@staticmethod
def filter_by_length(lengths: List[int], min_len: int, max_len: int):
def filter_by_length(lengths: list[int], min_len: int, max_len: int):
idxs = np.argsort(lengths) # ascending order
ignore_idx = []
keep_idx = []
@ -327,10 +374,9 @@ class TTSDataset(Dataset):
return ignore_idx, keep_idx
@staticmethod
def sort_by_length(samples: List[List]):
def sort_by_length(samples: list[list]):
audio_lengths = [s["audio_length"] for s in samples]
idxs = np.argsort(audio_lengths) # ascending order
return idxs
return np.argsort(audio_lengths) # ascending order
@staticmethod
def create_buckets(samples, batch_group_size: int):
@ -350,7 +396,7 @@ class TTSDataset(Dataset):
samples_new.append(samples[idx])
return samples_new
def preprocess_samples(self):
def preprocess_samples(self) -> None:
r"""Sort `items` based on text length or audio length in ascending order. Filter out samples out or the length
range.
"""
@ -376,7 +422,8 @@ class TTSDataset(Dataset):
samples = self._select_samples_by_idx(sorted_idxs, samples)
if len(samples) == 0:
raise RuntimeError(" [!] No samples left")
msg = "No samples left."
raise RuntimeError(msg)
# shuffle batch groups
# create batches with similar length items
@ -389,39 +436,38 @@ class TTSDataset(Dataset):
text_lengths = [s["text_length"] for s in samples]
self.samples = samples
if self.verbose:
print(" | > Preprocessing samples")
print(" | > Max text length: {}".format(np.max(text_lengths)))
print(" | > Min text length: {}".format(np.min(text_lengths)))
print(" | > Avg text length: {}".format(np.mean(text_lengths)))
print(" | ")
print(" | > Max audio length: {}".format(np.max(audio_lengths)))
print(" | > Min audio length: {}".format(np.min(audio_lengths)))
print(" | > Avg audio length: {}".format(np.mean(audio_lengths)))
print(f" | > Num. instances discarded samples: {len(ignore_idx)}")
print(" | > Batch group size: {}.".format(self.batch_group_size))
logger.info("Preprocessing samples")
logger.info(f"Max text length: {np.max(text_lengths)}")
logger.info(f"Min text length: {np.min(text_lengths)}")
logger.info(f"Avg text length: {np.mean(text_lengths)}")
logger.info(f"Max audio length: {np.max(audio_lengths)}")
logger.info(f"Min audio length: {np.min(audio_lengths)}")
logger.info(f"Avg audio length: {np.mean(audio_lengths)}")
logger.info("Num. instances discarded samples: %d", len(ignore_idx))
logger.info(f"Batch group size: {self.batch_group_size}.")
@staticmethod
def _sort_batch(batch, text_lengths):
"""Sort the batch by the input text length for RNN efficiency.
Args:
----
batch (Dict): Batch returned by `__getitem__`.
text_lengths (List[int]): Lengths of the input character sequences.
"""
text_lengths, ids_sorted_decreasing = torch.sort(torch.LongTensor(text_lengths), dim=0, descending=True)
batch = [batch[idx] for idx in ids_sorted_decreasing]
return batch, text_lengths, ids_sorted_decreasing
def collate_fn(self, batch):
r"""
Perform preprocessing and create a final data batch:
"""Perform preprocessing and create a final data batch.
1. Sort batch instances by text-length
2. Convert Audio signal to features.
3. PAD sequences wrt r.
4. Load to Torch.
"""
# Puts each data field into a tensor with outer dimension batch size
if isinstance(batch[0], collections.abc.Mapping):
token_ids_lengths = np.array([len(d["token_ids"]) for d in batch])
@ -456,9 +502,11 @@ class TTSDataset(Dataset):
# lengths adjusted by the reduction factor
mel_lengths_adjusted = [
(
m.shape[1] + (self.outputs_per_step - (m.shape[1] % self.outputs_per_step))
if m.shape[1] % self.outputs_per_step
else m.shape[1]
)
for m in mel
]
@ -564,23 +612,18 @@ class TTSDataset(Dataset):
"audio_unique_names": batch["audio_unique_name"],
}
raise TypeError(
(
"batch must contain tensors, numbers, dicts or lists;\
found {}".format(
type(batch[0])
)
)
)
msg = f"batch must contain tensors, numbers, dicts or lists; found {type(batch[0])}"
raise TypeError(msg)
class PhonemeDataset(Dataset):
"""Phoneme Dataset for converting input text to phonemes and then token IDs
"""Phoneme Dataset for converting input text to phonemes and then token IDs.
At initialization, it pre-computes the phonemes under `cache_path` and loads them in training to reduce data
loading latency. If `cache_path` is already present, it skips the pre-computation.
Args:
----
samples (Union[List[List], List[Dict]]):
List of samples. Each sample is a list or a dict.
@ -592,15 +635,16 @@ class PhonemeDataset(Dataset):
precompute_num_workers (int):
Number of workers used for pre-computing the phonemes. Defaults to 0.
"""
def __init__(
self,
samples: Union[List[Dict], List[List]],
samples: Union[list[dict], list[list]],
tokenizer: "TTSTokenizer",
cache_path: str,
precompute_num_workers=0,
):
precompute_num_workers: int = 0,
) -> None:
self.samples = samples
self.tokenizer = tokenizer
self.cache_path = cache_path
@ -608,16 +652,16 @@ class PhonemeDataset(Dataset):
os.makedirs(cache_path)
self.precompute(precompute_num_workers)
def __getitem__(self, index):
def __getitem__(self, index) -> dict[str, Any]:
item = self.samples[index]
ids = self.compute_or_load(string2filename(item["audio_unique_name"]), item["text"], item["language"])
ph_hat = self.tokenizer.ids_to_text(ids)
return {"text": item["text"], "ph_hat": ph_hat, "token_ids": ids, "token_ids_len": len(ids)}
def __len__(self):
def __len__(self) -> int:
return len(self.samples)
def compute_or_load(self, file_name, text, language):
def compute_or_load(self, file_name: str, text: str, language: str) -> list[int]:
"""Compute phonemes for the given text.
If the phonemes are already cached, load them from cache.
@ -631,20 +675,24 @@ class PhonemeDataset(Dataset):
np.save(cache_path, ids)
return ids
def get_pad_id(self):
"""Get pad token ID for sequence padding"""
def get_pad_id(self) -> int:
"""Get pad token ID for sequence padding."""
return self.tokenizer.pad_id
def precompute(self, num_workers=1):
def precompute(self, num_workers: int = 1) -> None:
"""Precompute phonemes for all samples.
We use pytorch dataloader because we are lazy.
"""
print("[*] Pre-computing phonemes...")
logger.info("Pre-computing phonemes...")
with tqdm.tqdm(total=len(self)) as pbar:
batch_size = num_workers if num_workers > 0 else 1
dataloder = torch.utils.data.DataLoader(
batch_size=batch_size, dataset=self, shuffle=False, num_workers=num_workers, collate_fn=self.collate_fn
batch_size=batch_size,
dataset=self,
shuffle=False,
num_workers=num_workers,
collate_fn=self.collate_fn,
)
for _ in dataloder:
pbar.update(batch_size)
@ -662,20 +710,20 @@ class PhonemeDataset(Dataset):
def print_logs(self, level: int = 0) -> None:
indent = "\t" * level
print("\n")
print(f"{indent}> PhonemeDataset ")
print(f"{indent}| > Tokenizer:")
logger.info("%sPhonemeDataset", indent)
logger.info("%s| Tokenizer:", indent)
self.tokenizer.print_logs(level + 1)
print(f"{indent}| > Number of instances : {len(self.samples)}")
logger.info("%s| Number of instances : %d", indent, len(self.samples))
class F0Dataset:
"""F0 Dataset for computing F0 from wav files in CPU
"""F0 Dataset for computing F0 from wav files in CPU.
Pre-compute F0 values for all the samples at initialization if `cache_path` is not None or already present. It
also computes the mean and std of F0 values if `normalize_f0` is True.
Args:
----
samples (Union[List[List], List[Dict]]):
List of samples. Each sample is a list or a dict.
@ -691,21 +739,20 @@ class F0Dataset:
normalize_f0 (bool):
Whether to normalize F0 values by mean and std. Defaults to True.
"""
def __init__(
self,
samples: Union[List[List], List[Dict]],
samples: Union[list[list], list[dict]],
ap: "AudioProcessor",
audio_config=None, # pylint: disable=unused-argument
verbose=False,
cache_path: str = None,
precompute_num_workers=0,
normalize_f0=True,
):
cache_path: Optional[str] = None,
precompute_num_workers: int = 0,
normalize_f0: bool = True,
) -> None:
self.samples = samples
self.ap = ap
self.verbose = verbose
self.cache_path = cache_path
self.normalize_f0 = normalize_f0
self.pad_id = 0.0
@ -725,18 +772,22 @@ class F0Dataset:
f0 = self.normalize(f0)
return {"audio_unique_name": item["audio_unique_name"], "f0": f0}
def __len__(self):
def __len__(self) -> int:
return len(self.samples)
def precompute(self, num_workers=0):
print("[*] Pre-computing F0s...")
def precompute(self, num_workers: int = 0) -> None:
logger.info("Pre-computing F0s...")
with tqdm.tqdm(total=len(self)) as pbar:
batch_size = num_workers if num_workers > 0 else 1
# we do not normalize at preproessing
normalize_f0 = self.normalize_f0
self.normalize_f0 = False
dataloder = torch.utils.data.DataLoader(
batch_size=batch_size, dataset=self, shuffle=False, num_workers=num_workers, collate_fn=self.collate_fn
batch_size=batch_size,
dataset=self,
shuffle=False,
num_workers=num_workers,
collate_fn=self.collate_fn,
)
computed_data = []
for batch in dataloder:
@ -755,9 +806,8 @@ class F0Dataset:
return self.pad_id
@staticmethod
def create_pitch_file_path(file_name, cache_path):
pitch_file = os.path.join(cache_path, file_name + "_pitch.npy")
return pitch_file
def create_pitch_file_path(file_name: str, cache_path: str) -> str:
return os.path.join(cache_path, file_name + "_pitch.npy")
@staticmethod
def _compute_and_save_pitch(ap, wav_file, pitch_file=None):
@ -773,7 +823,7 @@ class F0Dataset:
mean, std = np.mean(nonzeros), np.std(nonzeros)
return mean, std
def load_stats(self, cache_path):
def load_stats(self, cache_path) -> None:
stats_path = os.path.join(cache_path, "pitch_stats.npy")
stats = np.load(stats_path, allow_pickle=True).item()
self.mean = stats["mean"].astype(np.float32)
@ -794,9 +844,7 @@ class F0Dataset:
return pitch
def compute_or_load(self, wav_file, audio_unique_name):
"""
compute pitch and return a numpy array of pitch values
"""
"""Compute pitch and return a numpy array of pitch values."""
pitch_file = self.create_pitch_file_path(audio_unique_name, self.cache_path)
if not os.path.exists(pitch_file):
pitch = self._compute_and_save_pitch(self.ap, wav_file, pitch_file)
@ -816,18 +864,18 @@ class F0Dataset:
def print_logs(self, level: int = 0) -> None:
indent = "\t" * level
print("\n")
print(f"{indent}> F0Dataset ")
print(f"{indent}| > Number of instances : {len(self.samples)}")
logger.info("%sF0Dataset", indent)
logger.info("%s| Number of instances : %d", indent, len(self.samples))
class EnergyDataset:
"""Energy Dataset for computing Energy from wav files in CPU
"""Energy Dataset for computing Energy from wav files in CPU.
Pre-compute Energy values for all the samples at initialization if `cache_path` is not None or already present. It
also computes the mean and std of Energy values if `normalize_Energy` is True.
Args:
----
samples (Union[List[List], List[Dict]]):
List of samples. Each sample is a list or a dict.
@ -843,20 +891,19 @@ class EnergyDataset:
normalize_Energy (bool):
Whether to normalize Energy values by mean and std. Defaults to True.
"""
def __init__(
self,
samples: Union[List[List], List[Dict]],
samples: Union[list[list], list[dict]],
ap: "AudioProcessor",
verbose=False,
cache_path: str = None,
cache_path: Optional[str] = None,
precompute_num_workers=0,
normalize_energy=True,
):
) -> None:
self.samples = samples
self.ap = ap
self.verbose = verbose
self.cache_path = cache_path
self.normalize_energy = normalize_energy
self.pad_id = 0.0
@ -876,18 +923,22 @@ class EnergyDataset:
energy = self.normalize(energy)
return {"audio_unique_name": item["audio_unique_name"], "energy": energy}
def __len__(self):
def __len__(self) -> int:
return len(self.samples)
def precompute(self, num_workers=0):
print("[*] Pre-computing energys...")
def precompute(self, num_workers=0) -> None:
logger.info("Pre-computing energys...")
with tqdm.tqdm(total=len(self)) as pbar:
batch_size = num_workers if num_workers > 0 else 1
# we do not normalize at preproessing
normalize_energy = self.normalize_energy
self.normalize_energy = False
dataloder = torch.utils.data.DataLoader(
batch_size=batch_size, dataset=self, shuffle=False, num_workers=num_workers, collate_fn=self.collate_fn
batch_size=batch_size,
dataset=self,
shuffle=False,
num_workers=num_workers,
collate_fn=self.collate_fn,
)
computed_data = []
for batch in dataloder:
@ -908,8 +959,7 @@ class EnergyDataset:
@staticmethod
def create_energy_file_path(wav_file, cache_path):
file_name = os.path.splitext(os.path.basename(wav_file))[0]
energy_file = os.path.join(cache_path, file_name + "_energy.npy")
return energy_file
return os.path.join(cache_path, file_name + "_energy.npy")
@staticmethod
def _compute_and_save_energy(ap, wav_file, energy_file=None):
@ -925,7 +975,7 @@ class EnergyDataset:
mean, std = np.mean(nonzeros), np.std(nonzeros)
return mean, std
def load_stats(self, cache_path):
def load_stats(self, cache_path) -> None:
stats_path = os.path.join(cache_path, "energy_stats.npy")
stats = np.load(stats_path, allow_pickle=True).item()
self.mean = stats["mean"].astype(np.float32)
@ -946,9 +996,7 @@ class EnergyDataset:
return energy
def compute_or_load(self, wav_file, audio_unique_name):
"""
compute energy and return a numpy array of energy values
"""
"""Compute energy and return a numpy array of energy values."""
energy_file = self.create_energy_file_path(audio_unique_name, self.cache_path)
if not os.path.exists(energy_file):
energy = self._compute_and_save_energy(self.ap, wav_file, energy_file)
@ -968,6 +1016,5 @@ class EnergyDataset:
def print_logs(self, level: int = 0) -> None:
indent = "\t" * level
print("\n")
print(f"{indent}> energyDataset ")
print(f"{indent}| > Number of instances : {len(self.samples)}")
logger.info("%senergyDataset")
logger.info("%s| Number of instances : %d", indent, len(self.samples))

67
TTS/tts/datasets/formatters.py
View File

@ -1,3 +1,5 @@
import csv
import logging
import os
import re
import xml.etree.ElementTree as ET
@ -5,9 +7,10 @@ from glob import glob
from pathlib import Path
from typing import List
import pandas as pd
from tqdm import tqdm
logger = logging.getLogger(__name__)
########################
# DATASETS
########################
@ -23,32 +26,34 @@ def cml_tts(root_path, meta_file, ignored_speakers=None):
num_cols = len(lines[0].split("|")) # take the first row as reference
for idx, line in enumerate(lines[1:]):
if len(line.split("|")) != num_cols:
print(f" > Missing column in line {idx + 1} -> {line.strip()}")
logger.warning("Missing column in line %d -> %s", idx + 1, line.strip())
# load metadata
metadata = pd.read_csv(os.path.join(root_path, meta_file), sep="|")
assert all(x in metadata.columns for x in ["wav_filename", "transcript"])
client_id = None if "client_id" in metadata.columns else "default"
emotion_name = None if "emotion_name" in metadata.columns else "neutral"
with open(Path(root_path) / meta_file, newline="", encoding="utf-8") as f:
reader = csv.DictReader(f, delimiter="|")
metadata = list(reader)
assert all(x in metadata[0] for x in ["wav_filename", "transcript"])
client_id = None if "client_id" in metadata[0] else "default"
emotion_name = None if "emotion_name" in metadata[0] else "neutral"
items = []
not_found_counter = 0
for row in metadata.itertuples():
if client_id is None and ignored_speakers is not None and row.client_id in ignored_speakers:
for row in metadata:
if client_id is None and ignored_speakers is not None and row["client_id"] in ignored_speakers:
continue
audio_path = os.path.join(root_path, row.wav_filename)
audio_path = os.path.join(root_path, row["wav_filename"])
if not os.path.exists(audio_path):
not_found_counter += 1
continue
items.append(
{
"text": row.transcript,
"text": row["transcript"],
"audio_file": audio_path,
"speaker_name": client_id if client_id is not None else row.client_id,
"emotion_name": emotion_name if emotion_name is not None else row.emotion_name,
"speaker_name": client_id if client_id is not None else row["client_id"],
"emotion_name": emotion_name if emotion_name is not None else row["emotion_name"],
"root_path": root_path,
}
)
if not_found_counter > 0:
print(f" | > [!] {not_found_counter} files not found")
logger.warning("%d files not found", not_found_counter)
return items
@ -61,32 +66,34 @@ def coqui(root_path, meta_file, ignored_speakers=None):
num_cols = len(lines[0].split("|")) # take the first row as reference
for idx, line in enumerate(lines[1:]):
if len(line.split("|")) != num_cols:
print(f" > Missing column in line {idx + 1} -> {line.strip()}")
logger.warning("Missing column in line %d -> %s", idx + 1, line.strip())
# load metadata
metadata = pd.read_csv(os.path.join(root_path, meta_file), sep="|")
assert all(x in metadata.columns for x in ["audio_file", "text"])
speaker_name = None if "speaker_name" in metadata.columns else "coqui"
emotion_name = None if "emotion_name" in metadata.columns else "neutral"
with open(Path(root_path) / meta_file, newline="", encoding="utf-8") as f:
reader = csv.DictReader(f, delimiter="|")
metadata = list(reader)
assert all(x in metadata[0] for x in ["audio_file", "text"])
speaker_name = None if "speaker_name" in metadata[0] else "coqui"
emotion_name = None if "emotion_name" in metadata[0] else "neutral"
items = []
not_found_counter = 0
for row in metadata.itertuples():
if speaker_name is None and ignored_speakers is not None and row.speaker_name in ignored_speakers:
for row in metadata:
if speaker_name is None and ignored_speakers is not None and row["speaker_name"] in ignored_speakers:
continue
audio_path = os.path.join(root_path, row.audio_file)
audio_path = os.path.join(root_path, row["audio_file"])
if not os.path.exists(audio_path):
not_found_counter += 1
continue
items.append(
{
"text": row.text,
"text": row["text"],
"audio_file": audio_path,
"speaker_name": speaker_name if speaker_name is not None else row.speaker_name,
"emotion_name": emotion_name if emotion_name is not None else row.emotion_name,
"speaker_name": speaker_name if speaker_name is not None else row["speaker_name"],
"emotion_name": emotion_name if emotion_name is not None else row["emotion_name"],
"root_path": root_path,
}
)
if not_found_counter > 0:
print(f" | > [!] {not_found_counter} files not found")
logger.warning("%d files not found", not_found_counter)
return items
@ -169,7 +176,7 @@ def mailabs(root_path, meta_files=None, ignored_speakers=None):
if isinstance(ignored_speakers, list):
if speaker_name in ignored_speakers:
continue
print(" | > {}".format(csv_file))
logger.info(csv_file)
with open(txt_file, "r", encoding="utf-8") as ttf:
for line in ttf:
cols = line.split("|")
@ -184,7 +191,7 @@ def mailabs(root_path, meta_files=None, ignored_speakers=None):
)
else:
# M-AI-Labs have some missing samples, so just print the warning
print("> File %s does not exist!" % (wav_file))
logger.warning("File %s does not exist!", wav_file)
return items
@ -249,7 +256,7 @@ def sam_accenture(root_path, meta_file, **kwargs): # pylint: disable=unused-arg
text = item.text
wav_file = os.path.join(root_path, "vo_voice_quality_transformation", item.get("id") + ".wav")
if not os.path.exists(wav_file):
print(f" [!] {wav_file} in metafile does not exist. Skipping...")
logger.warning("%s in metafile does not exist. Skipping...", wav_file)
continue
items.append({"text": text, "audio_file": wav_file, "speaker_name": speaker_name, "root_path": root_path})
return items
@ -370,7 +377,7 @@ def custom_turkish(root_path, meta_file, **kwargs): # pylint: disable=unused-ar
continue
text = cols[1].strip()
items.append({"text": text, "audio_file": wav_file, "speaker_name": speaker_name, "root_path": root_path})
print(f" [!] {len(skipped_files)} files skipped. They don't exist...")
logger.warning("%d files skipped. They don't exist...")
return items
@ -438,7 +445,7 @@ def vctk(root_path, meta_files=None, wavs_path="wav48_silence_trimmed", mic="mic
{"text": text, "audio_file": wav_file, "speaker_name": "VCTK_" + speaker_id, "root_path": root_path}
)
else:
print(f" [!] wav files don't exist - {wav_file}")
logger.warning("Wav file doesn't exist - %s", wav_file)
return items

11
TTS/tts/layers/bark/hubert/hubert_manager.py
View File

@ -1,11 +1,14 @@
# From https://github.com/gitmylo/bark-voice-cloning-HuBERT-quantizer
import logging
import os.path
import shutil
import urllib.request
import huggingface_hub
logger = logging.getLogger(__name__)
class HubertManager:
@staticmethod
@ -13,9 +16,9 @@ class HubertManager:
download_url: str = "https://dl.fbaipublicfiles.com/hubert/hubert_base_ls960.pt", model_path: str = ""
):
if not os.path.isfile(model_path):
print("Downloading HuBERT base model")
logger.info("Downloading HuBERT base model")
urllib.request.urlretrieve(download_url, model_path)
print("Downloaded HuBERT")
logger.info("Downloaded HuBERT")
return model_path
return None
@ -27,9 +30,9 @@ class HubertManager:
):
model_dir = os.path.dirname(model_path)
if not os.path.isfile(model_path):
print("Downloading HuBERT custom tokenizer")
logger.info("Downloading HuBERT custom tokenizer")
huggingface_hub.hf_hub_download(repo, model, local_dir=model_dir, local_dir_use_symlinks=False)
shutil.move(os.path.join(model_dir, model), model_path)
print("Downloaded tokenizer")
logger.info("Downloaded tokenizer")
return model_path
return None

14
TTS/tts/layers/bark/hubert/kmeans_hubert.py
View File

@ -7,8 +7,6 @@ License: MIT
# Modified code from https://github.com/lucidrains/audiolm-pytorch/blob/main/audiolm_pytorch/hubert_kmeans.py
import logging
from pathlib import Path
import torch
from einops import pack, unpack
@ -16,6 +14,8 @@ from torch import nn
from torchaudio.functional import resample
from transformers import HubertModel
from TTS.utils.generic_utils import exists
def round_down_nearest_multiple(num, divisor):
return num // divisor * divisor
@ -28,21 +28,13 @@ def curtail_to_multiple(t, mult, from_left=False):
return t[..., seq_slice]
def exists(val):
return val is not None
def default(val, d):
return val if exists(val) else d
class CustomHubert(nn.Module):
"""
checkpoint and kmeans can be downloaded at https://github.com/facebookresearch/fairseq/tree/main/examples/hubert
or you can train your own
"""
def __init__(self, checkpoint_path, target_sample_hz=16000, seq_len_multiple_of=None, output_layer=9, device=None):
def __init__(self, target_sample_hz=16000, seq_len_multiple_of=None, output_layer=9, device=None):
super().__init__()
self.target_sample_hz = target_sample_hz
self.seq_len_multiple_of = seq_len_multiple_of

11
TTS/tts/layers/bark/hubert/tokenizer.py
View File

@ -5,6 +5,7 @@ License: MIT
"""
import json
import logging
import os.path
from zipfile import ZipFile
@ -12,6 +13,8 @@ import numpy
import torch
from torch import nn, optim
logger = logging.getLogger(__name__)
class HubertTokenizer(nn.Module):
def __init__(self, hidden_size=1024, input_size=768, output_size=10000, version=0):
@ -85,7 +88,7 @@ class HubertTokenizer(nn.Module):
# Print loss
if log_loss:
print("Loss", loss.item())
logger.info("Loss %.3f", loss.item())
# Backward pass
loss.backward()
@ -157,10 +160,10 @@ def auto_train(data_path, save_path="model.pth", load_model: str = None, save_ep
data_x, data_y = [], []
if load_model and os.path.isfile(load_model):
print("Loading model from", load_model)
logger.info("Loading model from %s", load_model)
model_training = HubertTokenizer.load_from_checkpoint(load_model, "cuda")
else:
print("Creating new model.")
logger.info("Creating new model.")
model_training = HubertTokenizer(version=1).to("cuda") # Settings for the model to run without lstm
save_path = os.path.join(data_path, save_path)
base_save_path = ".".join(save_path.split(".")[:-1])
@ -191,5 +194,5 @@ def auto_train(data_path, save_path="model.pth", load_model: str = None, save_ep
save_p_2 = f"{base_save_path}_epoch_{epoch}.pth"
model_training.save(save_p)
model_training.save(save_p_2)
print(f"Epoch {epoch} completed")
logger.info("Epoch %d completed", epoch)
epoch += 1

17
TTS/tts/layers/bark/inference_funcs.py
View File

@ -2,10 +2,11 @@ import logging
import os
import re
from glob import glob
from typing import Dict, List
from typing import Dict, List, Optional, Tuple
import librosa
import numpy as np
import numpy.typing as npt
import torch
import torchaudio
import tqdm
@ -48,7 +49,7 @@ def get_voices(extra_voice_dirs: List[str] = []): # pylint: disable=dangerous-d
return voices
def load_npz(npz_file):
def load_npz(npz_file: str) -> Tuple[npt.NDArray[np.int64], npt.NDArray[np.int64], npt.NDArray[np.int64]]:
x_history = np.load(npz_file)
semantic = x_history["semantic_prompt"]
coarse = x_history["coarse_prompt"]
@ -56,7 +57,11 @@ def load_npz(npz_file):
return semantic, coarse, fine
def load_voice(model, voice: str, extra_voice_dirs: List[str] = []): # pylint: disable=dangerous-default-value
def load_voice(
model, voice: str, extra_voice_dirs: List[str] = []
) -> Tuple[
Optional[npt.NDArray[np.int64]], Optional[npt.NDArray[np.int64]], Optional[npt.NDArray[np.int64]]
]: # pylint: disable=dangerous-default-value
if voice == "random":
return None, None, None
@ -107,11 +112,10 @@ def generate_voice(
model,
output_path,
):
"""Generate a new voice from a given audio and text prompt.
"""Generate a new voice from a given audio.
Args:
audio (np.ndarray): The audio to use as a base for the new voice.
text (str): Transcription of the audio you are clonning.
model (BarkModel): The BarkModel to use for generating the new voice.
output_path (str): The path to save the generated voice to.
"""
@ -130,10 +134,9 @@ def generate_voice(
# generate semantic tokens
# Load the HuBERT model
hubert_manager = HubertManager()
# hubert_manager.make_sure_hubert_installed(model_path=model.config.LOCAL_MODEL_PATHS["hubert"])
hubert_manager.make_sure_tokenizer_installed(model_path=model.config.LOCAL_MODEL_PATHS["hubert_tokenizer"])
hubert_model = CustomHubert(checkpoint_path=model.config.LOCAL_MODEL_PATHS["hubert"]).to(model.device)
hubert_model = CustomHubert().to(model.device)
# Load the CustomTokenizer model
tokenizer = HubertTokenizer.load_from_checkpoint(

12
TTS/tts/layers/bark/load_model.py
View File

@ -10,14 +10,10 @@ import tqdm
from TTS.tts.layers.bark.model import GPT, GPTConfig
from TTS.tts.layers.bark.model_fine import FineGPT, FineGPTConfig
from TTS.utils.generic_utils import is_pytorch_at_least_2_4
if (
torch.cuda.is_available()
and hasattr(torch.cuda, "amp")
and hasattr(torch.cuda.amp, "autocast")
and torch.cuda.is_bf16_supported()
):
autocast = functools.partial(torch.cuda.amp.autocast, dtype=torch.bfloat16)
if torch.cuda.is_available() and torch.cuda.is_bf16_supported():
autocast = functools.partial(torch.autocast, device_type="cuda", dtype=torch.bfloat16)
else:
@contextlib.contextmanager
@ -118,7 +114,7 @@ def load_model(ckpt_path, device, config, model_type="text"):
logger.info(f"{model_type} model not found, downloading...")
_download(config.REMOTE_MODEL_PATHS[model_type]["path"], ckpt_path, config.CACHE_DIR)
checkpoint = torch.load(ckpt_path, map_location=device)
checkpoint = torch.load(ckpt_path, map_location=device, weights_only=is_pytorch_at_least_2_4())
# this is a hack
model_args = checkpoint["model_args"]
if "input_vocab_size" not in model_args:

19
TTS/tts/layers/bark/model.py
View File

@ -2,6 +2,7 @@
Much of this code is adapted from Andrej Karpathy's NanoGPT
(https://github.com/karpathy/nanoGPT)
"""
import math
from dataclasses import dataclass
@ -11,18 +12,6 @@ from torch import nn
from torch.nn import functional as F
class LayerNorm(nn.Module):
"""LayerNorm but with an optional bias. PyTorch doesn't support simply bias=False"""
def __init__(self, ndim, bias):
super().__init__()
self.weight = nn.Parameter(torch.ones(ndim))
self.bias = nn.Parameter(torch.zeros(ndim)) if bias else None
def forward(self, x):
return F.layer_norm(x, self.weight.shape, self.weight, self.bias, 1e-5)
class CausalSelfAttention(nn.Module):
def __init__(self, config):
super().__init__()
@ -118,9 +107,9 @@ class MLP(nn.Module):
class Block(nn.Module):
def __init__(self, config, layer_idx):
super().__init__()
self.ln_1 = LayerNorm(config.n_embd, bias=config.bias)
self.ln_1 = nn.LayerNorm(config.n_embd, bias=config.bias)
self.attn = CausalSelfAttention(config)
self.ln_2 = LayerNorm(config.n_embd, bias=config.bias)
self.ln_2 = nn.LayerNorm(config.n_embd, bias=config.bias)
self.mlp = MLP(config)
self.layer_idx = layer_idx
@ -157,7 +146,7 @@ class GPT(nn.Module):
wpe=nn.Embedding(config.block_size, config.n_embd),
drop=nn.Dropout(config.dropout),
h=nn.ModuleList([Block(config, idx) for idx in range(config.n_layer)]),
ln_f=LayerNorm(config.n_embd, bias=config.bias),
ln_f=nn.LayerNorm(config.n_embd, bias=config.bias),
)
)
self.lm_head = nn.Linear(config.n_embd, config.output_vocab_size, bias=False)

1
TTS/tts/layers/bark/model_fine.py
View File

@ -2,6 +2,7 @@
Much of this code is adapted from Andrej Karpathy's NanoGPT
(https://github.com/karpathy/nanoGPT)
"""
import math
from dataclasses import dataclass

69
TTS/tts/layers/delightful_tts/acoustic_model.py
View File

@ -1,16 +1,17 @@
### credit: https://github.com/dunky11/voicesmith
import logging
from typing import Callable, Dict, Tuple
import torch
import torch.nn.functional as F
from coqpit import Coqpit
from monotonic_alignment_search import maximum_path
from torch import nn
from TTS.tts.layers.delightful_tts.conformer import Conformer
from TTS.tts.layers.delightful_tts.encoders import (
PhonemeLevelProsodyEncoder,
UtteranceLevelProsodyEncoder,
get_mask_from_lengths,
)
from TTS.tts.layers.delightful_tts.energy_adaptor import EnergyAdaptor
from TTS.tts.layers.delightful_tts.networks import EmbeddingPadded, positional_encoding
@ -18,7 +19,9 @@ from TTS.tts.layers.delightful_tts.phoneme_prosody_predictor import PhonemeProso
from TTS.tts.layers.delightful_tts.pitch_adaptor import PitchAdaptor
from TTS.tts.layers.delightful_tts.variance_predictor import VariancePredictor
from TTS.tts.layers.generic.aligner import AlignmentNetwork
from TTS.tts.utils.helpers import generate_path, maximum_path, sequence_mask
from TTS.tts.utils.helpers import expand_encoder_outputs, generate_attention, sequence_mask
logger = logging.getLogger(__name__)
class AcousticModel(torch.nn.Module):
@ -217,7 +220,7 @@ class AcousticModel(torch.nn.Module):
def _init_speaker_embedding(self):
# pylint: disable=attribute-defined-outside-init
if self.num_speakers > 0:
print(" > initialization of speaker-embedding layers.")
logger.info("Initialization of speaker-embedding layers.")
self.embedded_speaker_dim = self.args.speaker_embedding_channels
self.emb_g = nn.Embedding(self.num_speakers, self.embedded_speaker_dim)
@ -227,42 +230,6 @@ class AcousticModel(torch.nn.Module):
raise ValueError("[!] Speaker embedding layer already initialized before d_vector settings.")
self.embedded_speaker_dim = self.args.d_vector_dim
@staticmethod
def generate_attn(dr, x_mask, y_mask=None):
"""Generate an attention mask from the linear scale durations.
Args:
dr (Tensor): Linear scale durations.
x_mask (Tensor): Mask for the input (character) sequence.
y_mask (Tensor): Mask for the output (spectrogram) sequence. Compute it from the predicted durations
if None. Defaults to None.
Shapes
- dr: :math:`(B, T_{en})`
- x_mask: :math:`(B, T_{en})`
- y_mask: :math:`(B, T_{de})`
"""
# compute decode mask from the durations
if y_mask is None:
y_lengths = dr.sum(1).long()
y_lengths[y_lengths < 1] = 1
y_mask = torch.unsqueeze(sequence_mask(y_lengths, None), 1).to(dr.dtype)
attn_mask = torch.unsqueeze(x_mask, -1) * torch.unsqueeze(y_mask, 2)
attn = generate_path(dr, attn_mask.squeeze(1)).to(dr.dtype)
return attn
def _expand_encoder_with_durations(
self,
o_en: torch.FloatTensor,
dr: torch.IntTensor,
x_mask: torch.IntTensor,
y_lengths: torch.IntTensor,
):
y_mask = torch.unsqueeze(sequence_mask(y_lengths, None), 1).to(o_en.dtype)
attn = self.generate_attn(dr, x_mask, y_mask)
o_en_ex = torch.einsum("kmn, kjm -> kjn", [attn.float(), o_en])
return y_mask, o_en_ex, attn.transpose(1, 2)
def _forward_aligner(
self,
x: torch.FloatTensor,
@ -336,8 +303,8 @@ class AcousticModel(torch.nn.Module):
{"d_vectors": d_vectors, "speaker_ids": speaker_idx}
) # pylint: disable=unused-variable
src_mask = get_mask_from_lengths(src_lens) # [B, T_src]
mel_mask = get_mask_from_lengths(mel_lens) # [B, T_mel]
src_mask = ~sequence_mask(src_lens) # [B, T_src]
mel_mask = ~sequence_mask(mel_lens) # [B, T_mel]
# Token embeddings
token_embeddings = self.src_word_emb(tokens) # [B, T_src, C_hidden]
@ -362,7 +329,7 @@ class AcousticModel(torch.nn.Module):
pos_encoding = positional_encoding(
self.emb_dim,
max(token_embeddings.shape[1], max(mel_lens)),
max(token_embeddings.shape[1], *mel_lens),
device=token_embeddings.device,
)
encoder_outputs = self.encoder(
@ -416,8 +383,8 @@ class AcousticModel(torch.nn.Module):
encoder_outputs = encoder_outputs.transpose(1, 2) + pitch_emb + energy_emb
log_duration_prediction = self.duration_predictor(x=encoder_outputs_res.detach(), mask=src_mask)
mel_pred_mask, encoder_outputs_ex, alignments = self._expand_encoder_with_durations(
o_en=encoder_outputs, y_lengths=mel_lens, dr=dr, x_mask=~src_mask[:, None]
encoder_outputs_ex, alignments, mel_pred_mask = expand_encoder_outputs(
encoder_outputs, y_lengths=mel_lens, duration=dr, x_mask=~src_mask[:, None]
)
x = self.decoder(
@ -431,7 +398,7 @@ class AcousticModel(torch.nn.Module):
dr = torch.log(dr + 1)
dr_pred = torch.exp(log_duration_prediction) - 1
alignments_dp = self.generate_attn(dr_pred, src_mask.unsqueeze(1), mel_pred_mask) # [B, T_max, T_max2']
alignments_dp = generate_attention(dr_pred, src_mask.unsqueeze(1), mel_pred_mask) # [B, T_max, T_max2']
return {
"model_outputs": x,
@ -444,7 +411,7 @@ class AcousticModel(torch.nn.Module):
"p_prosody_pred": p_prosody_pred,
"p_prosody_ref": p_prosody_ref,
"alignments_dp": alignments_dp,
"alignments": alignments, # [B, T_de, T_en]
"alignments": alignments.transpose(1, 2), # [B, T_de, T_en]
"aligner_soft": aligner_soft,
"aligner_mas": aligner_mas,
"aligner_durations": aligner_durations,
@ -465,7 +432,7 @@ class AcousticModel(torch.nn.Module):
pitch_transform: Callable = None,
energy_transform: Callable = None,
) -> torch.Tensor:
src_mask = get_mask_from_lengths(torch.tensor([tokens.shape[1]], dtype=torch.int64, device=tokens.device))
src_mask = ~sequence_mask(torch.tensor([tokens.shape[1]], dtype=torch.int64, device=tokens.device))
src_lens = torch.tensor(tokens.shape[1:2]).to(tokens.device) # pylint: disable=unused-variable
sid, g, lid, _ = self._set_cond_input( # pylint: disable=unused-variable
{"d_vectors": d_vectors, "speaker_ids": speaker_idx}
@ -532,11 +499,11 @@ class AcousticModel(torch.nn.Module):
duration_pred = torch.round(duration_pred) # -> [B, T_src]
mel_lens = duration_pred.sum(1) # -> [B,]
_, encoder_outputs_ex, alignments = self._expand_encoder_with_durations(
o_en=encoder_outputs, y_lengths=mel_lens, dr=duration_pred.squeeze(1), x_mask=~src_mask[:, None]
encoder_outputs_ex, alignments, _ = expand_encoder_outputs(
encoder_outputs, y_lengths=mel_lens, duration=duration_pred.squeeze(1), x_mask=~src_mask[:, None]
)
mel_mask = get_mask_from_lengths(
mel_mask = ~sequence_mask(
torch.tensor([encoder_outputs_ex.shape[2]], dtype=torch.int64, device=encoder_outputs_ex.device)
)
@ -553,7 +520,7 @@ class AcousticModel(torch.nn.Module):
x = self.to_mel(x)
outputs = {
"model_outputs": x,
"alignments": alignments,
"alignments": alignments.transpose(1, 2),
# "pitch": pitch_emb_pred,
"durations": duration_pred,
"pitch": pitch_pred,

49
TTS/tts/layers/delightful_tts/conformer.py
View File

@ -1,20 +1,14 @@
### credit: https://github.com/dunky11/voicesmith
import math
from typing import Tuple
import torch
import torch.nn as nn # pylint: disable=consider-using-from-import
import torch.nn.functional as F
from TTS.tts.layers.delightful_tts.conv_layers import Conv1dGLU, DepthWiseConv1d, PointwiseConv1d
from TTS.tts.layers.delightful_tts.conv_layers import Conv1dGLU, DepthWiseConv1d, PointwiseConv1d, calc_same_padding
from TTS.tts.layers.delightful_tts.networks import GLUActivation
def calc_same_padding(kernel_size: int) -> Tuple[int, int]:
pad = kernel_size // 2
return (pad, pad - (kernel_size + 1) % 2)
class Conformer(nn.Module):
def __init__(
self,
@ -322,7 +316,7 @@ class ConformerMultiHeadedSelfAttention(nn.Module):
value: torch.Tensor,
mask: torch.Tensor,
encoding: torch.Tensor,
) -> Tuple[torch.Tensor, torch.Tensor]:
) -> tuple[torch.Tensor, torch.Tensor]:
batch_size, seq_length, _ = key.size() # pylint: disable=unused-variable
encoding = encoding[:, : key.shape[1]]
encoding = encoding.repeat(batch_size, 1, 1)
@ -378,7 +372,7 @@ class RelativeMultiHeadAttention(nn.Module):
value: torch.Tensor,
pos_embedding: torch.Tensor,
mask: torch.Tensor,
) -> Tuple[torch.Tensor, torch.Tensor]:
) -> tuple[torch.Tensor, torch.Tensor]:
batch_size = query.shape[0]
query = self.query_proj(query).view(batch_size, -1, self.num_heads, self.d_head)
key = self.key_proj(key).view(batch_size, -1, self.num_heads, self.d_head).permute(0, 2, 1, 3)
@ -411,40 +405,3 @@ class RelativeMultiHeadAttention(nn.Module):
padded_pos_score = padded_pos_score.view(batch_size, num_heads, seq_length2 + 1, seq_length1)
pos_score = padded_pos_score[:, :, 1:].view_as(pos_score)
return pos_score
class MultiHeadAttention(nn.Module):
"""
input:
query --- [N, T_q, query_dim]
key --- [N, T_k, key_dim]
output:
out --- [N, T_q, num_units]
"""
def __init__(self, query_dim: int, key_dim: int, num_units: int, num_heads: int):
super().__init__()
self.num_units = num_units
self.num_heads = num_heads
self.key_dim = key_dim
self.W_query = nn.Linear(in_features=query_dim, out_features=num_units, bias=False)
self.W_key = nn.Linear(in_features=key_dim, out_features=num_units, bias=False)
self.W_value = nn.Linear(in_features=key_dim, out_features=num_units, bias=False)
def forward(self, query: torch.Tensor, key: torch.Tensor) -> torch.Tensor:
querys = self.W_query(query) # [N, T_q, num_units]
keys = self.W_key(key) # [N, T_k, num_units]
values = self.W_value(key)
split_size = self.num_units // self.num_heads
querys = torch.stack(torch.split(querys, split_size, dim=2), dim=0) # [h, N, T_q, num_units/h]
keys = torch.stack(torch.split(keys, split_size, dim=2), dim=0) # [h, N, T_k, num_units/h]
values = torch.stack(torch.split(values, split_size, dim=2), dim=0) # [h, N, T_k, num_units/h]
# score = softmax(QK^T / (d_k ** 0.5))
scores = torch.matmul(querys, keys.transpose(2, 3)) # [h, N, T_q, T_k]
scores = scores / (self.key_dim**0.5)
scores = F.softmax(scores, dim=3)
# out = score * V
out = torch.matmul(scores, values) # [h, N, T_q, num_units/h]
out = torch.cat(torch.split(out, 1, dim=0), dim=3).squeeze(0) # [N, T_q, num_units]
return out

142
TTS/tts/layers/delightful_tts/conv_layers.py
View File

@ -3,9 +3,6 @@ from typing import Tuple
import torch
import torch.nn as nn # pylint: disable=consider-using-from-import
import torch.nn.functional as F
from torch.nn.utils import parametrize
from TTS.tts.layers.delightful_tts.kernel_predictor import KernelPredictor
def calc_same_padding(kernel_size: int) -> Tuple[int, int]:
@ -530,142 +527,3 @@ class CoordConv2d(nn.modules.conv.Conv2d):
x = self.addcoords(x)
x = self.conv(x)
return x
class LVCBlock(torch.nn.Module):
"""the location-variable convolutions"""
def __init__( # pylint: disable=dangerous-default-value
self,
in_channels,
cond_channels,
stride,
dilations=[1, 3, 9, 27],
lReLU_slope=0.2,
conv_kernel_size=3,
cond_hop_length=256,
kpnet_hidden_channels=64,
kpnet_conv_size=3,
kpnet_dropout=0.0,
):
super().__init__()
self.cond_hop_length = cond_hop_length
self.conv_layers = len(dilations)
self.conv_kernel_size = conv_kernel_size
self.kernel_predictor = KernelPredictor(
cond_channels=cond_channels,
conv_in_channels=in_channels,
conv_out_channels=2 * in_channels,
conv_layers=len(dilations),
conv_kernel_size=conv_kernel_size,
kpnet_hidden_channels=kpnet_hidden_channels,
kpnet_conv_size=kpnet_conv_size,
kpnet_dropout=kpnet_dropout,
kpnet_nonlinear_activation_params={"negative_slope": lReLU_slope},
)
self.convt_pre = nn.Sequential(
nn.LeakyReLU(lReLU_slope),
nn.utils.parametrizations.weight_norm(
nn.ConvTranspose1d(
in_channels,
in_channels,
2 * stride,
stride=stride,
padding=stride // 2 + stride % 2,
output_padding=stride % 2,
)
),
)
self.conv_blocks = nn.ModuleList()
for dilation in dilations:
self.conv_blocks.append(
nn.Sequential(
nn.LeakyReLU(lReLU_slope),
nn.utils.parametrizations.weight_norm(
nn.Conv1d(
in_channels,
in_channels,
conv_kernel_size,
padding=dilation * (conv_kernel_size - 1) // 2,
dilation=dilation,
)
),
nn.LeakyReLU(lReLU_slope),
)
)
def forward(self, x, c):
"""forward propagation of the location-variable convolutions.
Args:
x (Tensor): the input sequence (batch, in_channels, in_length)
c (Tensor): the conditioning sequence (batch, cond_channels, cond_length)
Returns:
Tensor: the output sequence (batch, in_channels, in_length)
"""
_, in_channels, _ = x.shape # (B, c_g, L')
x = self.convt_pre(x) # (B, c_g, stride * L')
kernels, bias = self.kernel_predictor(c)
for i, conv in enumerate(self.conv_blocks):
output = conv(x) # (B, c_g, stride * L')
k = kernels[:, i, :, :, :, :] # (B, 2 * c_g, c_g, kernel_size, cond_length)
b = bias[:, i, :, :] # (B, 2 * c_g, cond_length)
output = self.location_variable_convolution(
output, k, b, hop_size=self.cond_hop_length
) # (B, 2 * c_g, stride * L'): LVC
x = x + torch.sigmoid(output[:, :in_channels, :]) * torch.tanh(
output[:, in_channels:, :]
) # (B, c_g, stride * L'): GAU
return x
def location_variable_convolution(self, x, kernel, bias, dilation=1, hop_size=256): # pylint: disable=no-self-use
"""perform location-variable convolution operation on the input sequence (x) using the local convolution kernl.
Time: 414 μs ± 309 ns per loop (mean ± std. dev. of 7 runs, 1000 loops each), test on NVIDIA V100.
Args:
x (Tensor): the input sequence (batch, in_channels, in_length).
kernel (Tensor): the local convolution kernel (batch, in_channel, out_channels, kernel_size, kernel_length)
bias (Tensor): the bias for the local convolution (batch, out_channels, kernel_length)
dilation (int): the dilation of convolution.
hop_size (int): the hop_size of the conditioning sequence.
Returns:
(Tensor): the output sequence after performing local convolution. (batch, out_channels, in_length).
"""
batch, _, in_length = x.shape
batch, _, out_channels, kernel_size, kernel_length = kernel.shape
assert in_length == (kernel_length * hop_size), "length of (x, kernel) is not matched"
padding = dilation * int((kernel_size - 1) / 2)
x = F.pad(x, (padding, padding), "constant", 0) # (batch, in_channels, in_length + 2*padding)
x = x.unfold(2, hop_size + 2 * padding, hop_size) # (batch, in_channels, kernel_length, hop_size + 2*padding)
if hop_size < dilation:
x = F.pad(x, (0, dilation), "constant", 0)
x = x.unfold(
3, dilation, dilation
) # (batch, in_channels, kernel_length, (hop_size + 2*padding)/dilation, dilation)
x = x[:, :, :, :, :hop_size]
x = x.transpose(3, 4) # (batch, in_channels, kernel_length, dilation, (hop_size + 2*padding)/dilation)
x = x.unfold(4, kernel_size, 1) # (batch, in_channels, kernel_length, dilation, _, kernel_size)
o = torch.einsum("bildsk,biokl->bolsd", x, kernel)
o = o.to(memory_format=torch.channels_last_3d)
bias = bias.unsqueeze(-1).unsqueeze(-1).to(memory_format=torch.channels_last_3d)
o = o + bias
o = o.contiguous().view(batch, out_channels, -1)
return o
def remove_weight_norm(self):
self.kernel_predictor.remove_weight_norm()
parametrize.remove_parametrizations(self.convt_pre[1], "weight")
for block in self.conv_blocks:
parametrize.remove_parametrizations(block[1], "weight")

13
TTS/tts/layers/delightful_tts/encoders.py
View File

@ -7,14 +7,7 @@ import torch.nn.functional as F
from TTS.tts.layers.delightful_tts.conformer import ConformerMultiHeadedSelfAttention
from TTS.tts.layers.delightful_tts.conv_layers import CoordConv1d
from TTS.tts.layers.delightful_tts.networks import STL
def get_mask_from_lengths(lengths: torch.Tensor) -> torch.Tensor:
batch_size = lengths.shape[0]
max_len = torch.max(lengths).item()
ids = torch.arange(0, max_len, device=lengths.device).unsqueeze(0).expand(batch_size, -1)
mask = ids >= lengths.unsqueeze(1).expand(-1, max_len)
return mask
from TTS.tts.utils.helpers import sequence_mask
def stride_lens(lens: torch.Tensor, stride: int = 2) -> torch.Tensor:
@ -93,7 +86,7 @@ class ReferenceEncoder(nn.Module):
outputs --- [N, E//2]
"""
mel_masks = get_mask_from_lengths(mel_lens).unsqueeze(1)
mel_masks = ~sequence_mask(mel_lens).unsqueeze(1)
x = x.masked_fill(mel_masks, 0)
for conv, norm in zip(self.convs, self.norms):
x = conv(x)
@ -103,7 +96,7 @@ class ReferenceEncoder(nn.Module):
for _ in range(2):
mel_lens = stride_lens(mel_lens)
mel_masks = get_mask_from_lengths(mel_lens)
mel_masks = ~sequence_mask(mel_lens)
x = x.masked_fill(mel_masks.unsqueeze(1), 0)
x = x.permute((0, 2, 1))

128
TTS/tts/layers/delightful_tts/kernel_predictor.py
View File

@ -1,128 +0,0 @@
import torch.nn as nn # pylint: disable=consider-using-from-import
from torch.nn.utils import parametrize
class KernelPredictor(nn.Module):
"""Kernel predictor for the location-variable convolutions
Args:
cond_channels (int): number of channel for the conditioning sequence,
conv_in_channels (int): number of channel for the input sequence,
conv_out_channels (int): number of channel for the output sequence,
conv_layers (int): number of layers
"""
def __init__( # pylint: disable=dangerous-default-value
self,
cond_channels,
conv_in_channels,
conv_out_channels,
conv_layers,
conv_kernel_size=3,
kpnet_hidden_channels=64,
kpnet_conv_size=3,
kpnet_dropout=0.0,
kpnet_nonlinear_activation="LeakyReLU",
kpnet_nonlinear_activation_params={"negative_slope": 0.1},
):
super().__init__()
self.conv_in_channels = conv_in_channels
self.conv_out_channels = conv_out_channels
self.conv_kernel_size = conv_kernel_size
self.conv_layers = conv_layers
kpnet_kernel_channels = conv_in_channels * conv_out_channels * conv_kernel_size * conv_layers # l_w
kpnet_bias_channels = conv_out_channels * conv_layers # l_b
self.input_conv = nn.Sequential(
nn.utils.parametrizations.weight_norm(
nn.Conv1d(cond_channels, kpnet_hidden_channels, 5, padding=2, bias=True)
),
getattr(nn, kpnet_nonlinear_activation)(**kpnet_nonlinear_activation_params),
)
self.residual_convs = nn.ModuleList()
padding = (kpnet_conv_size - 1) // 2
for _ in range(3):
self.residual_convs.append(
nn.Sequential(
nn.Dropout(kpnet_dropout),
nn.utils.parametrizations.weight_norm(
nn.Conv1d(
kpnet_hidden_channels,
kpnet_hidden_channels,
kpnet_conv_size,
padding=padding,
bias=True,
)
),
getattr(nn, kpnet_nonlinear_activation)(**kpnet_nonlinear_activation_params),
nn.utils.parametrizations.weight_norm(
nn.Conv1d(
kpnet_hidden_channels,
kpnet_hidden_channels,
kpnet_conv_size,
padding=padding,
bias=True,
)
),
getattr(nn, kpnet_nonlinear_activation)(**kpnet_nonlinear_activation_params),
)
)
self.kernel_conv = nn.utils.parametrizations.weight_norm(
nn.Conv1d(
kpnet_hidden_channels,
kpnet_kernel_channels,
kpnet_conv_size,
padding=padding,
bias=True,
)
)
self.bias_conv = nn.utils.parametrizations.weight_norm(
nn.Conv1d(
kpnet_hidden_channels,
kpnet_bias_channels,
kpnet_conv_size,
padding=padding,
bias=True,
)
)
def forward(self, c):
"""
Args:
c (Tensor): the conditioning sequence (batch, cond_channels, cond_length)
"""
batch, _, cond_length = c.shape
c = self.input_conv(c)
for residual_conv in self.residual_convs:
residual_conv.to(c.device)
c = c + residual_conv(c)
k = self.kernel_conv(c)
b = self.bias_conv(c)
kernels = k.contiguous().view(
batch,
self.conv_layers,
self.conv_in_channels,
self.conv_out_channels,
self.conv_kernel_size,
cond_length,
)
bias = b.contiguous().view(
batch,
self.conv_layers,
self.conv_out_channels,
cond_length,
)
return kernels, bias
def remove_weight_norm(self):
parametrize.remove_parametrizations(self.input_conv[0], "weight")
parametrize.remove_parametrizations(self.kernel_conv, "weight")
parametrize.remove_parametrizations(self.bias_conv, "weight")
for block in self.residual_convs:
parametrize.remove_parametrizations(block[1], "weight")
parametrize.remove_parametrizations(block[3], "weight")

4
TTS/tts/layers/glow_tts/glow.py
View File

@ -1,5 +1,4 @@
import torch
from packaging.version import Version
from torch import nn
from torch.nn import functional as F
@ -90,9 +89,6 @@ class InvConvNear(nn.Module):
self.no_jacobian = no_jacobian
self.weight_inv = None
if Version(torch.__version__) < Version("1.9"):
w_init = torch.qr(torch.FloatTensor(self.num_splits, self.num_splits).normal_())[0]
else:
w_init = torch.linalg.qr(torch.FloatTensor(self.num_splits, self.num_splits).normal_(), "complete")[0]
if torch.det(w_init) < 0:

11
TTS/tts/layers/glow_tts/transformer.py
View File

@ -5,6 +5,7 @@ from torch import nn
from torch.nn import functional as F
from TTS.tts.layers.generic.normalization import LayerNorm, LayerNorm2
from TTS.tts.utils.helpers import convert_pad_shape
class RelativePositionMultiHeadAttention(nn.Module):
@ -300,7 +301,7 @@ class FeedForwardNetwork(nn.Module):
pad_l = self.kernel_size - 1
pad_r = 0
padding = [[0, 0], [0, 0], [pad_l, pad_r]]
x = F.pad(x, self._pad_shape(padding))
x = F.pad(x, convert_pad_shape(padding))
return x
def _same_padding(self, x):
@ -309,15 +310,9 @@ class FeedForwardNetwork(nn.Module):
pad_l = (self.kernel_size - 1) // 2
pad_r = self.kernel_size // 2
padding = [[0, 0], [0, 0], [pad_l, pad_r]]
x = F.pad(x, self._pad_shape(padding))
x = F.pad(x, convert_pad_shape(padding))
return x
@staticmethod
def _pad_shape(padding):
l = padding[::-1]
pad_shape = [item for sublist in l for item in sublist]
return pad_shape
class RelativePositionTransformer(nn.Module):
"""Transformer with Relative Potional Encoding.

96
TTS/tts/layers/losses.py
View File

@ -1,3 +1,4 @@
import logging
import math
import numpy as np
@ -10,6 +11,8 @@ from TTS.tts.utils.helpers import sequence_mask
from TTS.tts.utils.ssim import SSIMLoss as _SSIMLoss
from TTS.utils.audio.torch_transforms import TorchSTFT
logger = logging.getLogger(__name__)
# pylint: disable=abstract-method
# relates https://github.com/pytorch/pytorch/issues/42305
@ -132,11 +135,11 @@ class SSIMLoss(torch.nn.Module):
ssim_loss = self.loss_func((y_norm * mask).unsqueeze(1), (y_hat_norm * mask).unsqueeze(1))
if ssim_loss.item() > 1.0:
print(f" > SSIM loss is out-of-range {ssim_loss.item()}, setting it 1.0")
logger.info("SSIM loss is out-of-range (%.2f), setting it to 1.0", ssim_loss.item())
ssim_loss = torch.tensor(1.0, device=ssim_loss.device)
if ssim_loss.item() < 0.0:
print(f" > SSIM loss is out-of-range {ssim_loss.item()}, setting it 0.0")
logger.info("SSIM loss is out-of-range (%.2f), setting it to 0.0", ssim_loss.item())
ssim_loss = torch.tensor(0.0, device=ssim_loss.device)
return ssim_loss
@ -252,7 +255,7 @@ class GuidedAttentionLoss(torch.nn.Module):
@staticmethod
def _make_ga_mask(ilen, olen, sigma):
grid_x, grid_y = torch.meshgrid(torch.arange(olen).to(olen), torch.arange(ilen).to(ilen))
grid_x, grid_y = torch.meshgrid(torch.arange(olen).to(olen), torch.arange(ilen).to(ilen), indexing="ij")
grid_x, grid_y = grid_x.float(), grid_y.float()
return 1.0 - torch.exp(-((grid_y / ilen - grid_x / olen) ** 2) / (2 * (sigma**2)))
@ -306,6 +309,24 @@ class ForwardSumLoss(nn.Module):
return total_loss
class NLLLoss(nn.Module):
"""Negative log likelihood loss."""
def forward(self, log_prob: torch.Tensor) -> dict: # pylint: disable=no-self-use
"""Compute the loss.
Args:
logits (Tensor): [B, T, D]
Returns:
Tensor: [1]
"""
return_dict = {}
return_dict["loss"] = -log_prob.mean()
return return_dict
########################
# MODEL LOSS LAYERS
########################
@ -616,6 +637,28 @@ class AlignTTSLoss(nn.Module):
return {"loss": loss, "loss_l1": spec_loss, "loss_ssim": ssim_loss, "loss_dur": dur_loss, "mdn_loss": mdn_loss}
def feature_loss(feats_real, feats_generated):
loss = 0
for dr, dg in zip(feats_real, feats_generated):
for rl, gl in zip(dr, dg):
rl = rl.float().detach()
gl = gl.float()
loss += torch.mean(torch.abs(rl - gl))
return loss * 2
def generator_loss(scores_fake):
loss = 0
gen_losses = []
for dg in scores_fake:
dg = dg.float()
l = torch.mean((1 - dg) ** 2)
gen_losses.append(l)
loss += l
return loss, gen_losses
class VitsGeneratorLoss(nn.Module):
def __init__(self, c: Coqpit):
super().__init__()
@ -637,28 +680,6 @@ class VitsGeneratorLoss(nn.Module):
do_amp_to_db=True,
)
@staticmethod
def feature_loss(feats_real, feats_generated):
loss = 0
for dr, dg in zip(feats_real, feats_generated):
for rl, gl in zip(dr, dg):
rl = rl.float().detach()
gl = gl.float()
loss += torch.mean(torch.abs(rl - gl))
return loss * 2
@staticmethod
def generator_loss(scores_fake):
loss = 0
gen_losses = []
for dg in scores_fake:
dg = dg.float()
l = torch.mean((1 - dg) ** 2)
gen_losses.append(l)
loss += l
return loss, gen_losses
@staticmethod
def kl_loss(z_p, logs_q, m_p, logs_p, z_mask):
"""
@ -719,10 +740,8 @@ class VitsGeneratorLoss(nn.Module):
self.kl_loss(z_p=z_p, logs_q=logs_q, m_p=m_p, logs_p=logs_p, z_mask=z_mask.unsqueeze(1))
* self.kl_loss_alpha
)
loss_feat = (
self.feature_loss(feats_real=feats_disc_real, feats_generated=feats_disc_fake) * self.feat_loss_alpha
)
loss_gen = self.generator_loss(scores_fake=scores_disc_fake)[0] * self.gen_loss_alpha
loss_feat = feature_loss(feats_real=feats_disc_real, feats_generated=feats_disc_fake) * self.feat_loss_alpha
loss_gen = generator_loss(scores_fake=scores_disc_fake)[0] * self.gen_loss_alpha
loss_mel = torch.nn.functional.l1_loss(mel_slice, mel_slice_hat) * self.mel_loss_alpha
loss_duration = torch.sum(loss_duration.float()) * self.dur_loss_alpha
loss = loss_kl + loss_feat + loss_mel + loss_gen + loss_duration
@ -776,6 +795,15 @@ class VitsDiscriminatorLoss(nn.Module):
return return_dict
def _binary_alignment_loss(alignment_hard, alignment_soft):
"""Binary loss that forces soft alignments to match the hard alignments.
Explained in `https://arxiv.org/pdf/2108.10447.pdf`.
"""
log_sum = torch.log(torch.clamp(alignment_soft[alignment_hard == 1], min=1e-12)).sum()
return -log_sum / alignment_hard.sum()
class ForwardTTSLoss(nn.Module):
"""Generic configurable ForwardTTS loss."""
@ -817,14 +845,6 @@ class ForwardTTSLoss(nn.Module):
self.dur_loss_alpha = c.dur_loss_alpha
self.binary_alignment_loss_alpha = c.binary_align_loss_alpha
@staticmethod
def _binary_alignment_loss(alignment_hard, alignment_soft):
"""Binary loss that forces soft alignments to match the hard alignments as
explained in `https://arxiv.org/pdf/2108.10447.pdf`.
"""
log_sum = torch.log(torch.clamp(alignment_soft[alignment_hard == 1], min=1e-12)).sum()
return -log_sum / alignment_hard.sum()
def forward(
self,
decoder_output,
@ -876,7 +896,7 @@ class ForwardTTSLoss(nn.Module):
return_dict["loss_aligner"] = self.aligner_loss_alpha * aligner_loss
if self.binary_alignment_loss_alpha > 0 and alignment_hard is not None:
binary_alignment_loss = self._binary_alignment_loss(alignment_hard, alignment_soft)
binary_alignment_loss = _binary_alignment_loss(alignment_hard, alignment_soft)
loss = loss + self.binary_alignment_loss_alpha * binary_alignment_loss
if binary_loss_weight:
return_dict["loss_binary_alignment"] = (

7
TTS/tts/layers/overflow/common_layers.py
View File

@ -1,3 +1,4 @@
import logging
from typing import List, Tuple
import torch
@ -8,6 +9,8 @@ from tqdm.auto import tqdm
from TTS.tts.layers.tacotron.common_layers import Linear
from TTS.tts.layers.tacotron.tacotron2 import ConvBNBlock
logger = logging.getLogger(__name__)
class Encoder(nn.Module):
r"""Neural HMM Encoder
@ -213,8 +216,8 @@ class Outputnet(nn.Module):
original_tensor = std.clone().detach()
std = torch.clamp(std, min=self.std_floor)
if torch.any(original_tensor != std):
print(
"[*] Standard deviation was floored! The model is preventing overfitting, nothing serious to worry about"
logger.info(
"Standard deviation was floored! The model is preventing overfitting, nothing serious to worry about"
)
return std

3
TTS/tts/layers/overflow/neural_hmm.py
View File

@ -128,7 +128,8 @@ class NeuralHMM(nn.Module):
# Get mean, std and transition vector from decoder for this timestep
# Note: Gradient checkpointing currently doesn't works with multiple gpus inside a loop
if self.use_grad_checkpointing and self.training:
mean, std, transition_vector = checkpoint(self.output_net, h_memory, inputs)
# TODO: use_reentrant=False is recommended
mean, std, transition_vector = checkpoint(self.output_net, h_memory, inputs, use_reentrant=True)
else:
mean, std, transition_vector = self.output_net(h_memory, inputs)

2
TTS/tts/layers/overflow/plotting_utils.py
View File

@ -71,7 +71,7 @@ def plot_transition_probabilities_to_numpy(states, transition_probabilities, out
ax.set_title("Transition probability of state")
ax.set_xlabel("hidden state")
ax.set_ylabel("probability")
ax.set_xticks([i for i in range(len(transition_probabilities))]) # pylint: disable=unnecessary-comprehension
ax.set_xticks(list(range(len(transition_probabilities))))
ax.set_xticklabels([int(x) for x in states], rotation=90)
plt.tight_layout()
if not output_fig:

11
TTS/tts/layers/tacotron/capacitron_layers.py
View File

@ -3,6 +3,8 @@ from torch import nn
from torch.distributions.multivariate_normal import MultivariateNormal as MVN
from torch.nn import functional as F
from TTS.tts.layers.tacotron.common_layers import calculate_post_conv_height
class CapacitronVAE(nn.Module):
"""Effective Use of Variational Embedding Capacity for prosody transfer.
@ -97,7 +99,7 @@ class ReferenceEncoder(nn.Module):
self.training = False
self.bns = nn.ModuleList([nn.BatchNorm2d(num_features=filter_size) for filter_size in filters[1:]])
post_conv_height = self.calculate_post_conv_height(num_mel, 3, 2, 2, num_layers)
post_conv_height = calculate_post_conv_height(num_mel, 3, 2, 2, num_layers)
self.recurrence = nn.LSTM(
input_size=filters[-1] * post_conv_height, hidden_size=out_dim, batch_first=True, bidirectional=False
)
@ -155,13 +157,6 @@ class ReferenceEncoder(nn.Module):
return last_output.to(inputs.device) # [B, 128]
@staticmethod
def calculate_post_conv_height(height, kernel_size, stride, pad, n_convs):
"""Height of spec after n convolutions with fixed kernel/stride/pad."""
for _ in range(n_convs):
height = (height - kernel_size + 2 * pad) // stride + 1
return height
class TextSummary(nn.Module):
def __init__(self, embedding_dim, encoder_output_dim):

7
TTS/tts/layers/tacotron/common_layers.py
View File

@ -3,6 +3,13 @@ from torch import nn
from torch.nn import functional as F
def calculate_post_conv_height(height: int, kernel_size: int, stride: int, pad: int, n_convs: int) -> int:
"""Height of spec after n convolutions with fixed kernel/stride/pad."""
for _ in range(n_convs):
height = (height - kernel_size + 2 * pad) // stride + 1
return height
class Linear(nn.Module):
"""Linear layer with a specific initialization.

21
TTS/tts/layers/tacotron/gst_layers.py
View File

@ -2,6 +2,8 @@ import torch
import torch.nn.functional as F
from torch import nn
from TTS.tts.layers.tacotron.common_layers import calculate_post_conv_height
class GST(nn.Module):
"""Global Style Token Module for factorizing prosody in speech.
@ -44,7 +46,7 @@ class ReferenceEncoder(nn.Module):
self.convs = nn.ModuleList(convs)
self.bns = nn.ModuleList([nn.BatchNorm2d(num_features=filter_size) for filter_size in filters[1:]])
post_conv_height = self.calculate_post_conv_height(num_mel, 3, 2, 1, num_layers)
post_conv_height = calculate_post_conv_height(num_mel, 3, 2, 1, num_layers)
self.recurrence = nn.GRU(
input_size=filters[-1] * post_conv_height, hidden_size=embedding_dim // 2, batch_first=True
)
@ -71,13 +73,6 @@ class ReferenceEncoder(nn.Module):
return out.squeeze(0)
@staticmethod
def calculate_post_conv_height(height, kernel_size, stride, pad, n_convs):
"""Height of spec after n convolutions with fixed kernel/stride/pad."""
for _ in range(n_convs):
height = (height - kernel_size + 2 * pad) // stride + 1
return height
class StyleTokenLayer(nn.Module):
"""NN Module attending to style tokens based on prosody encodings."""
@ -117,7 +112,7 @@ class MultiHeadAttention(nn.Module):
out --- [N, T_q, num_units]
"""
def __init__(self, query_dim, key_dim, num_units, num_heads):
def __init__(self, query_dim: int, key_dim: int, num_units: int, num_heads: int):
super().__init__()
self.num_units = num_units
self.num_heads = num_heads
@ -127,7 +122,7 @@ class MultiHeadAttention(nn.Module):
self.W_key = nn.Linear(in_features=key_dim, out_features=num_units, bias=False)
self.W_value = nn.Linear(in_features=key_dim, out_features=num_units, bias=False)
def forward(self, query, key):
def forward(self, query: torch.Tensor, key: torch.Tensor) -> torch.Tensor:
queries = self.W_query(query) # [N, T_q, num_units]
keys = self.W_key(key) # [N, T_k, num_units]
values = self.W_value(key)
@ -137,13 +132,11 @@ class MultiHeadAttention(nn.Module):
keys = torch.stack(torch.split(keys, split_size, dim=2), dim=0) # [h, N, T_k, num_units/h]
values = torch.stack(torch.split(values, split_size, dim=2), dim=0) # [h, N, T_k, num_units/h]
# score = softmax(QK^T / (d_k**0.5))
# score = softmax(QK^T / (d_k ** 0.5))
scores = torch.matmul(queries, keys.transpose(2, 3)) # [h, N, T_q, T_k]
scores = scores / (self.key_dim**0.5)
scores = F.softmax(scores, dim=3)
# out = score * V
out = torch.matmul(scores, values) # [h, N, T_q, num_units/h]
out = torch.cat(torch.split(out, 1, dim=0), dim=3).squeeze(0) # [N, T_q, num_units]
return out
return torch.cat(torch.split(out, 1, dim=0), dim=3).squeeze(0) # [N, T_q, num_units]

6
TTS/tts/layers/tacotron/tacotron.py
View File

@ -1,12 +1,16 @@
# coding: utf-8
# adapted from https://github.com/r9y9/tacotron_pytorch
import logging
import torch
from torch import nn
from .attentions import init_attn
from .common_layers import Prenet
logger = logging.getLogger(__name__)
class BatchNormConv1d(nn.Module):
r"""A wrapper for Conv1d with BatchNorm. It sets the activation
@ -480,7 +484,7 @@ class Decoder(nn.Module):
if t > inputs.shape[1] / 4 and (stop_token > 0.6 or attention[:, -1].item() > 0.6):
break
if t > self.max_decoder_steps:
print(" | > Decoder stopped with 'max_decoder_steps")
logger.info("Decoder stopped with `max_decoder_steps` %d", self.max_decoder_steps)
break
return self._parse_outputs(outputs, attentions, stop_tokens)

8
TTS/tts/layers/tacotron/tacotron2.py
View File

@ -1,3 +1,5 @@
import logging
import torch
from torch import nn
from torch.nn import functional as F
@ -5,6 +7,8 @@ from torch.nn import functional as F
from .attentions import init_attn
from .common_layers import Linear, Prenet
logger = logging.getLogger(__name__)
# pylint: disable=no-value-for-parameter
# pylint: disable=unexpected-keyword-arg
@ -356,7 +360,7 @@ class Decoder(nn.Module):
if stop_token > self.stop_threshold and t > inputs.shape[0] // 2:
break
if len(outputs) == self.max_decoder_steps:
print(f" > Decoder stopped with `max_decoder_steps` {self.max_decoder_steps}")
logger.info("Decoder stopped with `max_decoder_steps` %d", self.max_decoder_steps)
break
memory = self._update_memory(decoder_output)
@ -389,7 +393,7 @@ class Decoder(nn.Module):
if stop_token > 0.7:
break
if len(outputs) == self.max_decoder_steps:
print(" | > Decoder stopped with 'max_decoder_steps")
logger.info("Decoder stopped with `max_decoder_steps` %d", self.max_decoder_steps)
break
self.memory_truncated = decoder_output

131
TTS/tts/layers/tortoise/arch_utils.py
View File

@ -1,6 +1,5 @@
import functools
import math
import os
import fsspec
import torch
@ -10,6 +9,7 @@ import torchaudio
from transformers import LogitsWarper
from TTS.tts.layers.tortoise.xtransformers import ContinuousTransformerWrapper, RelativePositionBias
from TTS.utils.generic_utils import is_pytorch_at_least_2_4
def zero_module(module):
@ -70,11 +70,10 @@ class QKVAttentionLegacy(nn.Module):
weight = rel_pos(weight.reshape(bs, self.n_heads, weight.shape[-2], weight.shape[-1])).reshape(
bs * self.n_heads, weight.shape[-2], weight.shape[-1]
)
weight = torch.softmax(weight.float(), dim=-1).type(weight.dtype)
if mask is not None:
# The proper way to do this is to mask before the softmax using -inf, but that doesn't work properly on CPUs.
mask = mask.repeat(self.n_heads, 1).unsqueeze(1)
weight = weight * mask
mask = mask.repeat(self.n_heads, 1, 1)
weight[mask.logical_not()] = -torch.inf
weight = torch.softmax(weight.float(), dim=-1).type(weight.dtype)
a = torch.einsum("bts,bcs->bct", weight, v)
return a.reshape(bs, -1, length)
@ -93,12 +92,12 @@ class AttentionBlock(nn.Module):
channels,
num_heads=1,
num_head_channels=-1,
do_checkpoint=True,
*,
relative_pos_embeddings=False,
tortoise_norm=False,
):
super().__init__()
self.channels = channels
self.do_checkpoint = do_checkpoint
if num_head_channels == -1:
self.num_heads = num_heads
else:
@ -110,6 +109,7 @@ class AttentionBlock(nn.Module):
self.qkv = nn.Conv1d(channels, channels * 3, 1)
# split heads before split qkv
self.attention = QKVAttentionLegacy(self.num_heads)
self.tortoise_norm = tortoise_norm
self.proj_out = zero_module(nn.Conv1d(channels, channels, 1))
if relative_pos_embeddings:
@ -126,10 +126,13 @@ class AttentionBlock(nn.Module):
def forward(self, x, mask=None):
b, c, *spatial = x.shape
x = x.reshape(b, c, -1)
qkv = self.qkv(self.norm(x))
x_norm = self.norm(x)
qkv = self.qkv(x_norm)
h = self.attention(qkv, mask, self.relative_pos_embeddings)
h = self.proj_out(h)
if self.tortoise_norm:
return (x + h).reshape(b, c, *spatial)
return (x_norm + h).reshape(b, c, *spatial)
class Upsample(nn.Module):
@ -185,115 +188,7 @@ class Downsample(nn.Module):
return self.op(x)
class ResBlock(nn.Module):
def __init__(
self,
channels,
dropout,
out_channels=None,
use_conv=False,
use_scale_shift_norm=False,
up=False,
down=False,
kernel_size=3,
):
super().__init__()
self.channels = channels
self.dropout = dropout
self.out_channels = out_channels or channels
self.use_conv = use_conv
self.use_scale_shift_norm = use_scale_shift_norm
padding = 1 if kernel_size == 3 else 2
self.in_layers = nn.Sequential(
normalization(channels),
nn.SiLU(),
nn.Conv1d(channels, self.out_channels, kernel_size, padding=padding),
)
self.updown = up or down
if up:
self.h_upd = Upsample(channels, False)
self.x_upd = Upsample(channels, False)
elif down:
self.h_upd = Downsample(channels, False)
self.x_upd = Downsample(channels, False)
else:
self.h_upd = self.x_upd = nn.Identity()
self.out_layers = nn.Sequential(
normalization(self.out_channels),
nn.SiLU(),
nn.Dropout(p=dropout),
zero_module(nn.Conv1d(self.out_channels, self.out_channels, kernel_size, padding=padding)),
)
if self.out_channels == channels:
self.skip_connection = nn.Identity()
elif use_conv:
self.skip_connection = nn.Conv1d(channels, self.out_channels, kernel_size, padding=padding)
else:
self.skip_connection = nn.Conv1d(channels, self.out_channels, 1)
def forward(self, x):
if self.updown:
in_rest, in_conv = self.in_layers[:-1], self.in_layers[-1]
h = in_rest(x)
h = self.h_upd(h)
x = self.x_upd(x)
h = in_conv(h)
else:
h = self.in_layers(x)
h = self.out_layers(h)
return self.skip_connection(x) + h
class AudioMiniEncoder(nn.Module):
def __init__(
self,
spec_dim,
embedding_dim,
base_channels=128,
depth=2,
resnet_blocks=2,
attn_blocks=4,
num_attn_heads=4,
dropout=0,
downsample_factor=2,
kernel_size=3,
):
super().__init__()
self.init = nn.Sequential(nn.Conv1d(spec_dim, base_channels, 3, padding=1))
ch = base_channels
res = []
for l in range(depth):
for r in range(resnet_blocks):
res.append(ResBlock(ch, dropout, kernel_size=kernel_size))
res.append(Downsample(ch, use_conv=True, out_channels=ch * 2, factor=downsample_factor))
ch *= 2
self.res = nn.Sequential(*res)
self.final = nn.Sequential(normalization(ch), nn.SiLU(), nn.Conv1d(ch, embedding_dim, 1))
attn = []
for a in range(attn_blocks):
attn.append(
AttentionBlock(
embedding_dim,
num_attn_heads,
)
)
self.attn = nn.Sequential(*attn)
self.dim = embedding_dim
def forward(self, x):
h = self.init(x)
h = self.res(h)
h = self.final(h)
h = self.attn(h)
return h[:, :, 0]
DEFAULT_MEL_NORM_FILE = "https://coqui.gateway.scarf.sh/v0.14.1_models/mel_norms.pth"
DEFAULT_MEL_NORM_FILE = "https://github.com/coqui-ai/TTS/releases/download/v0.14.1_models/mel_norms.pth"
class TorchMelSpectrogram(nn.Module):
@ -333,7 +228,7 @@ class TorchMelSpectrogram(nn.Module):
self.mel_norm_file = mel_norm_file
if self.mel_norm_file is not None:
with fsspec.open(self.mel_norm_file) as f:
self.mel_norms = torch.load(f)
self.mel_norms = torch.load(f, weights_only=is_pytorch_at_least_2_4())
else:
self.mel_norms = None

32
TTS/tts/layers/tortoise/audio_utils.py
View File

@ -1,3 +1,4 @@
import logging
import os
from glob import glob
from typing import Dict, List
@ -8,7 +9,10 @@ import torch
import torchaudio
from scipy.io.wavfile import read
from TTS.utils.audio.torch_transforms import TorchSTFT
from TTS.utils.audio.torch_transforms import TorchSTFT, amp_to_db
from TTS.utils.generic_utils import is_pytorch_at_least_2_4
logger = logging.getLogger(__name__)
def load_wav_to_torch(full_path):
@ -28,7 +32,7 @@ def check_audio(audio, audiopath: str):
# Check some assumptions about audio range. This should be automatically fixed in load_wav_to_torch, but might not be in some edge cases, where we should squawk.
# '2' is arbitrarily chosen since it seems like audio will often "overdrive" the [-1,1] bounds.
if torch.any(audio > 2) or not torch.any(audio < 0):
print(f"Error with {audiopath}. Max={audio.max()} min={audio.min()}")
logger.error("Error with %s. Max=%.2f min=%.2f", audiopath, audio.max(), audio.min())
audio.clip_(-1, 1)
@ -84,24 +88,6 @@ def normalize_tacotron_mel(mel):
return 2 * ((mel - TACOTRON_MEL_MIN) / (TACOTRON_MEL_MAX - TACOTRON_MEL_MIN)) - 1
def dynamic_range_compression(x, C=1, clip_val=1e-5):
"""
PARAMS
------
C: compression factor
"""
return torch.log(torch.clamp(x, min=clip_val) * C)
def dynamic_range_decompression(x, C=1):
"""
PARAMS
------
C: compression factor used to compress
"""
return torch.exp(x) / C
def get_voices(extra_voice_dirs: List[str] = []):
dirs = extra_voice_dirs
voices: Dict[str, List[str]] = {}
@ -121,7 +107,7 @@ def load_voice(voice: str, extra_voice_dirs: List[str] = []):
voices = get_voices(extra_voice_dirs)
paths = voices[voice]
if len(paths) == 1 and paths[0].endswith(".pth"):
return None, torch.load(paths[0])
return None, torch.load(paths[0], weights_only=is_pytorch_at_least_2_4())
else:
conds = []
for cond_path in paths:
@ -136,7 +122,7 @@ def load_voices(voices: List[str], extra_voice_dirs: List[str] = []):
for voice in voices:
if voice == "random":
if len(voices) > 1:
print("Cannot combine a random voice with a non-random voice. Just using a random voice.")
logger.warning("Cannot combine a random voice with a non-random voice. Just using a random voice.")
return None, None
clip, latent = load_voice(voice, extra_voice_dirs)
if latent is None:
@ -171,7 +157,7 @@ def wav_to_univnet_mel(wav, do_normalization=False, device="cuda"):
)
stft = stft.to(device)
mel = stft(wav)
mel = dynamic_range_compression(mel)
mel = amp_to_db(mel)
if do_normalization:
mel = normalize_tacotron_mel(mel)
return mel

112
TTS/tts/layers/tortoise/autoregressive.py
View File

@ -1,14 +1,23 @@
# AGPL: a notification must be added stating that changes have been made to that file.
import functools
import random
from typing import Optional
import torch
import torch.nn as nn
import torch.nn.functional as F
import transformers
from packaging.version import Version
from transformers import GPT2Config, GPT2PreTrainedModel, LogitsProcessorList
from transformers.modeling_outputs import CausalLMOutputWithCrossAttentions
from TTS.tts.layers.tortoise.arch_utils import AttentionBlock, TypicalLogitsWarper
if Version(transformers.__version__) >= Version("4.45"):
isin = transformers.pytorch_utils.isin_mps_friendly
else:
isin = torch.isin
def null_position_embeddings(range, dim):
return torch.zeros((range.shape[0], range.shape[1], dim), device=range.device)
@ -115,7 +124,7 @@ class GPT2InferenceModel(GPT2PreTrainedModel):
else:
emb = self.embeddings(input_ids)
emb = emb + self.text_pos_embedding.get_fixed_embedding(
attention_mask.shape[1] - mel_len, attention_mask.device
attention_mask.shape[1] - (mel_len + 1), attention_mask.device
)
transformer_outputs = self.transformer(
@ -167,44 +176,56 @@ class ConditioningEncoder(nn.Module):
embedding_dim,
attn_blocks=6,
num_attn_heads=4,
do_checkpointing=False,
mean=False,
*,
tortoise_norm=False,
):
super().__init__()
attn = []
self.init = nn.Conv1d(spec_dim, embedding_dim, kernel_size=1)
for a in range(attn_blocks):
attn.append(AttentionBlock(embedding_dim, num_attn_heads))
attn.append(AttentionBlock(embedding_dim, num_attn_heads, tortoise_norm=tortoise_norm))
self.attn = nn.Sequential(*attn)
self.dim = embedding_dim
self.do_checkpointing = do_checkpointing
self.mean = mean
def forward(self, x):
"""
x: (b, 80, s)
"""
h = self.init(x)
h = self.attn(h)
if self.mean:
return h.mean(dim=2)
else:
return h[:, :, 0]
return h
class LearnedPositionEmbeddings(nn.Module):
def __init__(self, seq_len, model_dim, init=0.02):
def __init__(self, seq_len, model_dim, init=0.02, relative=False):
super().__init__()
self.emb = nn.Embedding(seq_len, model_dim)
# Initializing this way is standard for GPT-2
self.emb.weight.data.normal_(mean=0.0, std=init)
self.relative = relative
self.seq_len = seq_len
def forward(self, x):
sl = x.shape[1]
if self.relative:
start = random.randint(sl, self.seq_len) - sl
return self.emb(torch.arange(start, start + sl, device=x.device))
else:
return self.emb(torch.arange(0, sl, device=x.device))
def get_fixed_embedding(self, ind, dev):
return self.emb(torch.arange(0, ind, device=dev))[ind - 1 : ind]
return self.emb(torch.tensor([ind], device=dev)).unsqueeze(0)
def build_hf_gpt_transformer(layers, model_dim, heads, max_mel_seq_len, max_text_seq_len, checkpointing):
def build_hf_gpt_transformer(
layers: int,
model_dim: int,
heads: int,
max_mel_seq_len: int,
max_text_seq_len: int,
checkpointing: bool,
max_prompt_len: int = 0,
):
"""
GPT-2 implemented by the HuggingFace library.
"""
@ -212,8 +233,8 @@ def build_hf_gpt_transformer(layers, model_dim, heads, max_mel_seq_len, max_text
gpt_config = GPT2Config(
vocab_size=256, # Unused.
n_positions=max_mel_seq_len + max_text_seq_len,
n_ctx=max_mel_seq_len + max_text_seq_len,
n_positions=max_mel_seq_len + max_text_seq_len + max_prompt_len,
n_ctx=max_mel_seq_len + max_text_seq_len + max_prompt_len,
n_embd=model_dim,
n_layer=layers,
n_head=heads,
@ -226,13 +247,18 @@ def build_hf_gpt_transformer(layers, model_dim, heads, max_mel_seq_len, max_text
gpt.wpe = functools.partial(null_position_embeddings, dim=model_dim)
# Built-in token embeddings are unused.
del gpt.wte
return (
gpt,
LearnedPositionEmbeddings(max_mel_seq_len, model_dim),
LearnedPositionEmbeddings(max_text_seq_len, model_dim),
None,
None,
mel_pos_emb = (
LearnedPositionEmbeddings(max_mel_seq_len, model_dim)
if max_mel_seq_len != -1
else functools.partial(null_position_embeddings, dim=model_dim)
)
text_pos_emb = (
LearnedPositionEmbeddings(max_text_seq_len, model_dim)
if max_mel_seq_len != -1
else functools.partial(null_position_embeddings, dim=model_dim)
)
return gpt, mel_pos_emb, text_pos_emb, None, None
class MelEncoder(nn.Module):
@ -326,12 +352,12 @@ class UnifiedVoice(nn.Module):
self.mel_layer_pos_embedding,
self.text_layer_pos_embedding,
) = build_hf_gpt_transformer(
layers,
model_dim,
heads,
self.max_mel_tokens + 2 + self.max_conditioning_inputs,
self.max_text_tokens + 2,
checkpointing,
layers=layers,
model_dim=model_dim,
heads=heads,
max_mel_seq_len=self.max_mel_tokens + 2 + self.max_conditioning_inputs,
max_text_seq_len=self.max_text_tokens + 2,
checkpointing=checkpointing,
)
if train_solo_embeddings:
self.mel_solo_embedding = nn.Parameter(torch.randn(1, 1, model_dim) * 0.02, requires_grad=True)
@ -447,7 +473,7 @@ class UnifiedVoice(nn.Module):
)
conds = []
for j in range(speech_conditioning_input.shape[1]):
conds.append(self.conditioning_encoder(speech_conditioning_input[:, j]))
conds.append(self.conditioning_encoder(speech_conditioning_input[:, j])[:, :, 0])
conds = torch.stack(conds, dim=1)
conds = conds.mean(dim=1)
return conds
@ -596,6 +622,8 @@ class UnifiedVoice(nn.Module):
max_length = (
trunc_index + self.max_mel_tokens - 1 if max_generate_length is None else trunc_index + max_generate_length
)
stop_token_tensor = torch.tensor(self.stop_mel_token, device=inputs.device, dtype=torch.long)
attention_mask = _prepare_attention_mask_for_generation(inputs, stop_token_tensor, stop_token_tensor)
gen = self.inference_model.generate(
inputs,
bos_token_id=self.start_mel_token,
@ -604,11 +632,39 @@ class UnifiedVoice(nn.Module):
max_length=max_length,
logits_processor=logits_processor,
num_return_sequences=num_return_sequences,
attention_mask=attention_mask,
**hf_generate_kwargs,
)
return gen[:, trunc_index:]
def _prepare_attention_mask_for_generation(
inputs: torch.Tensor,
pad_token_id: Optional[torch.Tensor],
eos_token_id: Optional[torch.Tensor],
) -> torch.LongTensor:
# No information for attention mask inference -> return default attention mask
default_attention_mask = torch.ones(inputs.shape[:2], dtype=torch.long, device=inputs.device)
if pad_token_id is None:
return default_attention_mask
is_input_ids = len(inputs.shape) == 2 and inputs.dtype in [torch.int, torch.long]
if not is_input_ids:
return default_attention_mask
is_pad_token_in_inputs = (pad_token_id is not None) and (isin(elements=inputs, test_elements=pad_token_id).any())
is_pad_token_not_equal_to_eos_token_id = (eos_token_id is None) or ~(
isin(elements=eos_token_id, test_elements=pad_token_id).any()
)
can_infer_attention_mask = is_pad_token_in_inputs * is_pad_token_not_equal_to_eos_token_id
attention_mask_from_padding = inputs.ne(pad_token_id).long()
attention_mask = (
attention_mask_from_padding * can_infer_attention_mask + default_attention_mask * ~can_infer_attention_mask
)
return attention_mask
if __name__ == "__main__":
gpt = UnifiedVoice(
model_dim=256,

6
TTS/tts/layers/tortoise/classifier.py
View File

@ -16,7 +16,6 @@ class ResBlock(nn.Module):
up=False,
down=False,
kernel_size=3,
do_checkpoint=True,
):
super().__init__()
self.channels = channels
@ -24,7 +23,6 @@ class ResBlock(nn.Module):
self.out_channels = out_channels or channels
self.use_conv = use_conv
self.use_scale_shift_norm = use_scale_shift_norm
self.do_checkpoint = do_checkpoint
padding = 1 if kernel_size == 3 else 2
self.in_layers = nn.Sequential(
@ -92,14 +90,14 @@ class AudioMiniEncoder(nn.Module):
self.layers = depth
for l in range(depth):
for r in range(resnet_blocks):
res.append(ResBlock(ch, dropout, do_checkpoint=False, kernel_size=kernel_size))
res.append(ResBlock(ch, dropout, kernel_size=kernel_size))
res.append(Downsample(ch, use_conv=True, out_channels=ch * 2, factor=downsample_factor))
ch *= 2
self.res = nn.Sequential(*res)
self.final = nn.Sequential(normalization(ch), nn.SiLU(), nn.Conv1d(ch, embedding_dim, 1))
attn = []
for a in range(attn_blocks):
attn.append(AttentionBlock(embedding_dim, num_attn_heads, do_checkpoint=False))
attn.append(AttentionBlock(embedding_dim, num_attn_heads, tortoise_norm=True))
self.attn = nn.Sequential(*attn)
self.dim = embedding_dim

Some files were not shown because too many files have changed in this diff Show More