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 pull request #3081 from coqui-ai/dev 

v0.17.9
This commit is contained in:
Eren Gölge 2023-10-19 11:23:55 +02:00 committed by GitHub
parent df2422eb72 bf68848f38
commit f0faed962d
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
17 changed files with 290 additions and 186 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
README.md
View File

@ -146,7 +146,7 @@ Underlined "TTS*" and "Judy*" are **internal** 🐸TTS models that are not relea
You can also help us implement more models.
## Installation
🐸TTS is tested on Ubuntu 18.04 with **python >= 3.7, < 3.11.**.
🐸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.
@ -198,17 +198,18 @@ from TTS.api import TTS
# Get device
device = "cuda" if torch.cuda.is_available() else "cpu"
# List available 🐸TTS models and choose the first one
model_name = TTS().list_models()[0]
# List available 🐸TTS models
print(TTS().list_models())
# Init TTS
tts = TTS(model_name).to(device)
tts = TTS("tts_models/multilingual/multi-dataset/xtts_v1").to(device)
# Run TTS
# ❗ Since this model is multi-speaker and multi-lingual, we must set the target speaker and the language
# Text to speech with a numpy output
wav = tts.tts("This is a test! This is also a test!!", speaker=tts.speakers[0], language=tts.languages[0])
# ❗ 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=tts.speakers[0], language=tts.languages[0], file_path="output.wav")
tts.tts_to_file(text="Hello world!", speaker_wav="my/cloning/audio.wav", language="en", file_path="output.wav")
```
#### Running a single speaker model
@ -347,6 +348,18 @@ If you don't specify any models, then it uses LJSpeech based English model.
$ 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
```
- Run TTS and define speed factor to use for 🐸Coqui Studio models, between 0.0 and 2.0:
```
$ tts --text "Text for TTS" --model_name "coqui_studio/<language>/<dataset>/<model_name>" --speed 1.2 --out_path output/path/speech.wav
```
- Run a TTS model with its default vocoder model:
```

6
TTS/.models.json
View File

@ -5,9 +5,9 @@
"xtts_v1": {
"description": "XTTS-v1 by Coqui with 13 languages and cross-language voice cloning.",
"hf_url": [
"https://huggingface.co/coqui/XTTS-v1/resolve/hifigan/model.pth",
"https://huggingface.co/coqui/XTTS-v1/resolve/hifigan/config.json",
"https://huggingface.co/coqui/XTTS-v1/resolve/hifigan/vocab.json"
"https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v1/hifigan/model.pth",
"https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v1/hifigan/config.json",
"https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v1/hifigan/vocab.json",
],
"default_vocoder": null,
"commit": "e5140314",

2
TTS/VERSION
View File

@ -1 +1 @@
0.17.8
0.17.9

18
TTS/api.py
View File

@ -112,7 +112,6 @@ class TTS(nn.Module):
return self.synthesizer.tts_model.language_manager.num_languages > 1
return False
@property
def speakers(self):
if not self.is_multi_speaker:
@ -265,6 +264,7 @@ class TTS(nn.Module):
language: str = None,
emotion: str = None,
speed: float = 1.0,
pipe_out = None,
file_path: str = None,
) -> Union[np.ndarray, str]:
"""Convert text to speech using Coqui Studio models. Use `CS_API` class if you are only interested in the API.
@ -281,6 +281,8 @@ class TTS(nn.Module):
with "V1" model. Defaults to None.
speed (float, optional):
Speed of the speech. Defaults to 1.0.
pipe_out (BytesIO, optional):
Flag to stdout the generated TTS wav file for shell pipe.
file_path (str, optional):
Path to save the output file. When None it returns the `np.ndarray` of waveform. Defaults to None.
@ -294,6 +296,7 @@ class TTS(nn.Module):
speaker_name=speaker_name,
language=language,
speed=speed,
pipe_out=pipe_out,
emotion=emotion,
file_path=file_path,
)[0]
@ -356,6 +359,7 @@ class TTS(nn.Module):
speaker_wav: str = None,
emotion: str = None,
speed: float = 1.0,
pipe_out = None,
file_path: str = "output.wav",
**kwargs,
):
@ -377,6 +381,8 @@ class TTS(nn.Module):
Emotion to use for 🐸Coqui Studio models. Defaults to "Neutral".
speed (float, optional):
Speed factor to use for 🐸Coqui Studio models, between 0.0 and 2.0. Defaults to None.
pipe_out (BytesIO, optional):
Flag to stdout the generated TTS wav file for shell pipe.
file_path (str, optional):
Output file path. Defaults to "output.wav".
kwargs (dict, optional):
@ -386,10 +392,16 @@ class TTS(nn.Module):
if self.csapi is not None:
return self.tts_coqui_studio(
text=text, speaker_name=speaker, language=language, emotion=emotion, speed=speed, file_path=file_path
text=text,
speaker_name=speaker,
language=language,
emotion=emotion,
speed=speed,
file_path=file_path,
pipe_out=pipe_out,
)
wav = self.tts(text=text, speaker=speaker, language=language, speaker_wav=speaker_wav, **kwargs)
self.synthesizer.save_wav(wav=wav, path=file_path)
self.synthesizer.save_wav(wav=wav, path=file_path, pipe_out=pipe_out)
return file_path
def voice_conversion(

41
TTS/bin/synthesize.py
View File

@ -2,6 +2,7 @@
# -*- coding: utf-8 -*-
import argparse
import contextlib
import sys
from argparse import RawTextHelpFormatter
@ -59,6 +60,18 @@ If you don't specify any models, then it uses LJSpeech based English model.
$ 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
```
- Run TTS and define speed factor to use for 🐸Coqui Studio models, between 0.0 and 2.0:
```
$ tts --text "Text for TTS" --model_name "coqui_studio/<language>/<dataset>/<model_name>" --speed 1.2 --out_path output/path/speech.wav
```
- Run a TTS model with its default vocoder model:
```
@ -228,6 +241,20 @@ def main():
help="Language to condition the model with. Only available for 🐸Coqui Studio `XTTS-multilingual` model.",
default=None,
)
parser.add_argument(
"--pipe_out",
help="stdout the generated TTS wav file for shell pipe.",
type=str2bool,
nargs="?",
const=True,
default=False,
)
parser.add_argument(
"--speed",
type=float,
help="Speed factor to use for 🐸Coqui Studio models, between 0.0 and 2.0.",
default=None,
)
# args for multi-speaker synthesis
parser.add_argument("--speakers_file_path", type=str, help="JSON file for multi-speaker model.", default=None)
@ -335,6 +362,9 @@ def main():
if not any(check_args):
parser.parse_args(["-h"])
pipe_out = sys.stdout if args.pipe_out else None
with contextlib.redirect_stdout(None if args.pipe_out else sys.stdout):
# Late-import to make things load faster
from TTS.api import TTS
from TTS.utils.manage import ModelManager
@ -378,7 +408,14 @@ def main():
if "coqui_studio" in args.model_name:
print(" > Using 🐸Coqui Studio model: ", args.model_name)
api = TTS(model_name=args.model_name, cs_api_model=args.cs_model)
api.tts_to_file(text=args.text, emotion=args.emotion, file_path=args.out_path, language=args.language)
api.tts_to_file(
text=args.text,
emotion=args.emotion,
file_path=args.out_path,
language=args.language,
speed=args.speed,
pipe_out=pipe_out,
)
print(" > Saving output to ", args.out_path)
return
@ -495,7 +532,7 @@ def main():
# save the results
print(" > Saving output to {}".format(args.out_path))
synthesizer.save_wav(wav, args.out_path)
synthesizer.save_wav(wav, args.out_path, pipe_out=pipe_out)
if __name__ == "__main__":

6
TTS/cs_api.py
View File

@ -9,6 +9,8 @@ import numpy as np
import requests
from scipy.io import wavfile
from TTS.utils.audio.numpy_transforms import save_wav
class Speaker(object):
"""Convert dict to object."""
@ -288,6 +290,7 @@ class CS_API:
speaker_id=None,
emotion=None,
speed=1.0,
pipe_out=None,
language=None,
file_path: str = None,
) -> str:
@ -300,6 +303,7 @@ class CS_API:
speaker_id (str): Speaker ID. If None, the speaker name is used.
emotion (str): Emotion of the speaker. One of "Neutral", "Happy", "Sad", "Angry", "Dull".
speed (float): Speed of the speech. 1.0 is normal speed.
pipe_out (BytesIO, optional): Flag to stdout the generated TTS wav file for shell pipe.
language (str): Language of the text. If None, the default language of the speaker is used. Language is only
supported by `XTTS-multilang` model. Currently supports en, de, es, fr, it, pt, pl. Defaults to "en".
file_path (str): Path to save the file. If None, a temporary file is created.
@ -307,7 +311,7 @@ class CS_API:
if file_path is None:
file_path = tempfile.mktemp(".wav")
wav, sr = self.tts(text, speaker_name, speaker_id, emotion, speed, language)
wavfile.write(file_path, sr, wav)
save_wav(wav=wav, path=file_path, sample_rate=sr, pipe_out=pipe_out)
return file_path

4
TTS/tts/models/forward_tts.py
View File

@ -396,6 +396,7 @@ class ForwardTTS(BaseTTS):
- g: :math:`(B, C)`
"""
if hasattr(self, "emb_g"):
g = g.type(torch.LongTensor)
g = self.emb_g(g) # [B, C, 1]
if g is not None:
g = g.unsqueeze(-1)
@ -683,9 +684,10 @@ class ForwardTTS(BaseTTS):
# encoder pass
o_en, x_mask, g, _ = self._forward_encoder(x, x_mask, g)
# duration predictor pass
o_dr_log = self.duration_predictor(o_en, x_mask)
o_dr_log = self.duration_predictor(o_en.squeeze(), x_mask)
o_dr = self.format_durations(o_dr_log, x_mask).squeeze(1)
y_lengths = o_dr.sum(1)
# pitch predictor pass
o_pitch = None
if self.args.use_pitch:

13
TTS/utils/audio/numpy_transforms.py
View File

@ -1,3 +1,4 @@
from io import BytesIO
from typing import Tuple
import librosa
@ -427,16 +428,24 @@ def load_wav(*, filename: str, sample_rate: int = None, resample: bool = False,
return x
def save_wav(*, wav: np.ndarray, path: str, sample_rate: int = None, **kwargs) -> None:
def save_wav(*, wav: np.ndarray, path: str, sample_rate: int = None, pipe_out = None, **kwargs) -> None:
"""Save float waveform to a file using Scipy.
Args:
wav (np.ndarray): Waveform with float values in range [-1, 1] to save.
path (str): Path to a output file.
sr (int, optional): Sampling rate used for saving to the file. Defaults to None.
pipe_out (BytesIO, optional): Flag to stdout the generated TTS wav file for shell pipe.
"""
wav_norm = wav * (32767 / max(0.01, np.max(np.abs(wav))))
scipy.io.wavfile.write(path, sample_rate, wav_norm.astype(np.int16))
wav_norm = wav_norm.astype(np.int16)
if pipe_out:
wav_buffer = BytesIO()
scipy.io.wavfile.write(wav_buffer, sample_rate, wav_norm)
wav_buffer.seek(0)
pipe_out.buffer.write(wav_buffer.read())
scipy.io.wavfile.write(path, sample_rate, wav_norm)
def mulaw_encode(*, wav: np.ndarray, mulaw_qc: int, **kwargs) -> np.ndarray:

12
TTS/utils/audio/processor.py
View File

@ -1,3 +1,4 @@
from io import BytesIO
from typing import Dict, Tuple
import librosa
@ -693,20 +694,27 @@ class AudioProcessor(object):
x = self.rms_volume_norm(x, self.db_level)
return x
def save_wav(self, wav: np.ndarray, path: str, sr: int = None) -> None:
def save_wav(self, wav: np.ndarray, path: str, sr: int = None, pipe_out = None) -> None:
"""Save a waveform to a file using Scipy.
Args:
wav (np.ndarray): Waveform to save.
path (str): Path to a output file.
sr (int, optional): Sampling rate used for saving to the file. Defaults to None.
pipe_out (BytesIO, optional): Flag to stdout the generated TTS wav file for shell pipe.
"""
if self.do_rms_norm:
wav_norm = self.rms_volume_norm(wav, self.db_level) * 32767
else:
wav_norm = wav * (32767 / max(0.01, np.max(np.abs(wav))))
scipy.io.wavfile.write(path, sr if sr else self.sample_rate, wav_norm.astype(np.int16))
wav_norm = wav_norm.astype(np.int16)
if pipe_out:
wav_buffer = BytesIO()
scipy.io.wavfile.write(wav_buffer, sr if sr else self.sample_rate, wav_norm)
wav_buffer.seek(0)
pipe_out.buffer.write(wav_buffer.read())
scipy.io.wavfile.write(path, sr if sr else self.sample_rate, wav_norm)
def get_duration(self, filename: str) -> float:
"""Get the duration of a wav file using Librosa.

15
TTS/utils/synthesizer.py
View File

@ -235,19 +235,20 @@ class Synthesizer(nn.Module):
"""
return self.seg.segment(text)
def save_wav(self, wav: List[int], path: str) -> None:
def save_wav(self, wav: List[int], path: str, pipe_out = None) -> None:
"""Save the waveform as a file.
Args:
wav (List[int]): waveform as a list of values.
path (str): output path to save the waveform.
pipe_out (BytesIO, optional): Flag to stdout the generated TTS wav file for shell pipe.
"""
# if tensor convert to numpy
if torch.is_tensor(wav):
wav = wav.cpu().numpy()
if isinstance(wav, list):
wav = np.array(wav)
save_wav(wav=wav, path=path, sample_rate=self.output_sample_rate)
save_wav(wav=wav, path=path, sample_rate=self.output_sample_rate, pipe_out=pipe_out)
def voice_conversion(self, source_wav: str, target_wav: str) -> List[int]:
output_wav = self.vc_model.voice_conversion(source_wav, target_wav)
@ -299,11 +300,7 @@ class Synthesizer(nn.Module):
speaker_embedding = None
speaker_id = None
if self.tts_speakers_file or hasattr(self.tts_model.speaker_manager, "name_to_id"):
# handle Neon models with single speaker.
if len(self.tts_model.speaker_manager.name_to_id) == 1:
speaker_id = list(self.tts_model.speaker_manager.name_to_id.values())[0]
elif speaker_name and isinstance(speaker_name, str):
if speaker_name and isinstance(speaker_name, str):
if self.tts_config.use_d_vector_file:
# get the average speaker embedding from the saved d_vectors.
speaker_embedding = self.tts_model.speaker_manager.get_mean_embedding(
@ -313,7 +310,9 @@ class Synthesizer(nn.Module):
else:
# get speaker idx from the speaker name
speaker_id = self.tts_model.speaker_manager.name_to_id[speaker_name]
# handle Neon models with single speaker.
elif len(self.tts_model.speaker_manager.name_to_id) == 1:
speaker_id = list(self.tts_model.speaker_manager.name_to_id.values())[0]
elif not speaker_name and not speaker_wav:
raise ValueError(
" [!] Looks like you are using a multi-speaker model. "

11
docs/source/formatting_your_dataset.md
View File

@ -17,19 +17,20 @@ Let's assume you created the audio clips and their transcription. You can collec
...
```
You can either create separate transcription files for each clip or create a text file that maps each audio clip to its transcription. In this file, each line must be delimitered by a special character separating the audio file name from the transcription. And make sure that the delimiter is not used in the transcription text.
You can either create separate transcription files for each clip or create a text file that maps each audio clip to its transcription. In this file, each column must be delimitered by a special character separating the audio file name, the transcription and the normalized transcription. And make sure that the delimiter is not used in the transcription text.
We recommend the following format delimited by `|`. In the following example, `audio1`, `audio2` refer to files `audio1.wav`, `audio2.wav` etc.
```
# metadata.txt
audio1|This is my sentence.
audio2|This is maybe my sentence.
audio3|This is certainly my sentence.
audio4|Let this be your sentence.
audio1|This is my sentence.|This is my sentence.
audio2|1469 and 1470|fourteen sixty-nine and fourteen seventy
audio3|It'll be $16 sir.|It'll be sixteen dollars sir.
...
```
*If you don't have normalized transcriptions, you can use the same transcription for both columns. If it's your case, we recommend to use normalization later in the pipeline, either in the text cleaner or in the phonemizer.*
In the end, we have the following folder structure
```

2
docs/source/implementing_a_new_model.md
View File

@ -41,7 +41,7 @@
6. Optionally, define `MyModelArgs`.
`MyModelArgs` is a 👨‍✈️Coqpit class that sets all the class arguments of the `MyModel`. `MyModelArgs` must have
all the fields neccessary to instantiate the `MyModel`. However, for training, you need to pass `MyModelConfig` to
all the fields necessary to instantiate the `MyModel`. However, for training, you need to pass `MyModelConfig` to
the model.
7. Test `MyModel`.

20
docs/source/inference.md
View File

@ -114,18 +114,24 @@ tts-server --model_name "<type>/<language>/<dataset>/<model_name>" \
You can run a multi-speaker and multi-lingual model in Python as
```python
import torch
from TTS.api import TTS
# List available 🐸TTS models and choose the first one
model_name = TTS().list_models()[0]
# Get device
device = "cuda" if torch.cuda.is_available() else "cpu"
# List available 🐸TTS models
print(TTS().list_models())
# Init TTS
tts = TTS(model_name)
tts = TTS("tts_models/multilingual/multi-dataset/xtts_v1").to(device)
# Run TTS
# ❗ Since this model is multi-speaker and multi-lingual, we must set the target speaker and the language
# Text to speech with a numpy output
wav = tts.tts("This is a test! This is also a test!!", speaker=tts.speakers[0], language=tts.languages[0])
# ❗ 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=tts.speakers[0], language=tts.languages[0], file_path="output.wav")
tts.tts_to_file(text="Hello world!", speaker_wav="my/cloning/audio.wav", language="en", file_path="output.wav")
```
#### Here is an example for a single speaker model.

2
docs/source/main_classes/trainer_api.md
View File

@ -1,3 +1,3 @@
# Trainer API
We made the trainer a seprate project on https://github.com/coqui-ai/Trainer
We made the trainer a separate project on https://github.com/coqui-ai/Trainer

2
docs/source/models/forward_tts.md
View File

@ -12,7 +12,7 @@ Currently we provide the following pre-configured architectures:
- **FastPitch:**
It uses the same FastSpeech architecture that is conditioned on fundemental frequency (f0) contours with the
It uses the same FastSpeech architecture that is conditioned on fundamental frequency (f0) contours with the
promise of more expressive speech.
- **SpeedySpeech:**

2
notebooks/dataset_analysis/AnalyzeDataset.ipynb
View File

@ -100,7 +100,7 @@
" wav_file = item[\"audio_file\"].strip()\n",
" wav_files.append(wav_file)\n",
" if not os.path.exists(wav_file):\n",
" print(waf_path)"
" print(wav_file)"
]
},
{

13
tests/api_tests/test_synthesize_api.py
View File

@ -13,3 +13,16 @@ def test_synthesize():
'--text "This is it" '
f'--out_path "{output_path}"'
)
# 🐸 Coqui studio model with speed arg.
run_cli(
'tts --model_name "coqui_studio/en/Torcull Diarmuid/coqui_studio" '
'--text "This is it but slow" --speed 0.1'
f'--out_path "{output_path}"'
)
# test pipe_out command
run_cli(
'tts --text "test." --pipe_out '
f'--out_path "{output_path}" | aplay'
)