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 remote-tracking branch 'TTS/dev' into dev

This commit is contained in:
thllwg 2020-07-02 13:35:48 +02:00
parent 1a0d8c1558 b1935c97fa
commit c96028f464
84 changed files with 5143 additions and 646 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
.cardboardlint.yml
View File

@ -1,2 +1,5 @@
linters: linters:
- pylint: - pylint:
# pylintrc: pylintrc
filefilter: ['- test_*.py', '+ *.py', '- *.npy']
# exclude:

3
.github/ISSUE_TEMPLATE.md vendored
View File

@ -6,8 +6,9 @@ labels: ''
assignees: '' assignees: ''
--- ---
<b>Questions</b> will not be answered here!!
Please consider posting on [TTS Discourse](https://discourse.mozilla.org/c/tts) page, if your issue is not directly related to TTS development process. Please consider posting on [TTS Discourse](https://discourse.mozilla.org/c/tts) page if your issue is not directly related to TTS development (Bugs, code updates etc.).
You can also check https://github.com/mozilla/TTS/wiki/FAQ for common questions and answers. You can also check https://github.com/mozilla/TTS/wiki/FAQ for common questions and answers.

3
.pylintrc
View File

@ -157,7 +157,8 @@ disable=missing-docstring,
xreadlines-attribute, xreadlines-attribute,
deprecated-sys-function, deprecated-sys-function,
exception-escape, exception-escape,
comprehension-escape comprehension-escape,
duplicate-code
# Enable the message, report, category or checker with the given id(s). You can # 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 # either give multiple identifier separated by comma (,) or put this option

4
.travis.yml
View File

@ -3,6 +3,10 @@ language: python
git: git:
quiet: true quiet: true
before_install:
- sudo apt-get update
- sudo apt-get -y install espeak
matrix: matrix:
include: include:
- name: "Lint check" - name: "Lint check"

4
.travis/script
View File

@ -12,7 +12,9 @@ fi
if [[ "$TEST_SUITE" == "unittest" ]]; then if [[ "$TEST_SUITE" == "unittest" ]]; then
# Run tests on all pushes # Run tests on all pushes
pushd tts_namespace pushd tts_namespace
python -m unittest nosetests TTS.speaker_encoder.tests --nocapture
nosetests TTS.vocoder.tests --nocapture
nosetests TTS.tests --nocapture
popd popd
# Test server package # Test server package
./tests/test_server_package.sh ./tests/test_server_package.sh

64
README.md
View File

@ -3,25 +3,49 @@
<img src="https://travis-ci.org/mozilla/TTS.svg?branch=dev"/> <img src="https://travis-ci.org/mozilla/TTS.svg?branch=dev"/>
This project is a part of [Mozilla Common Voice](https://voice.mozilla.org/en). TTS aims a deep learning based Text2Speech engine, low in cost and high in quality. To begin with, you can hear a sample generated voice from [here](https://soundcloud.com/user-565970875/commonvoice-loc-sens-attn). This project is a part of [Mozilla Common Voice](https://voice.mozilla.org/en). TTS aims a deep learning based Text2Speech engine, low in cost and high in quality. To begin with, you can hear a sample synthesized voice from [here](https://soundcloud.com/user-565970875/commonvoice-loc-sens-attn).
TTS includes two different model implementations which are based on [Tacotron](https://arxiv.org/abs/1703.10135) and [Tacotron2](https://arxiv.org/abs/1712.05884). Tacotron is smaller, efficient and easier to train but Tacotron2 provides better results, especially when it is combined with a Neural vocoder. Therefore, choose depending on your project requirements.
If you are new, you can also find [here](http://www.erogol.com/text-speech-deep-learning-architectures/) a brief post about TTS architectures and their comparisons. If you are new, you can also find [here](http://www.erogol.com/text-speech-deep-learning-architectures/) a brief post about TTS architectures and their comparisons.
[![](https://sourcerer.io/fame/erogol/erogol/TTS/images/0)](https://sourcerer.io/fame/erogol/erogol/TTS/links/0)[![](https://sourcerer.io/fame/erogol/erogol/TTS/images/1)](https://sourcerer.io/fame/erogol/erogol/TTS/links/1)[![](https://sourcerer.io/fame/erogol/erogol/TTS/images/2)](https://sourcerer.io/fame/erogol/erogol/TTS/links/2)[![](https://sourcerer.io/fame/erogol/erogol/TTS/images/3)](https://sourcerer.io/fame/erogol/erogol/TTS/links/3)[![](https://sourcerer.io/fame/erogol/erogol/TTS/images/4)](https://sourcerer.io/fame/erogol/erogol/TTS/links/4)[![](https://sourcerer.io/fame/erogol/erogol/TTS/images/5)](https://sourcerer.io/fame/erogol/erogol/TTS/links/5)[![](https://sourcerer.io/fame/erogol/erogol/TTS/images/6)](https://sourcerer.io/fame/erogol/erogol/TTS/links/6)[![](https://sourcerer.io/fame/erogol/erogol/TTS/images/7)](https://sourcerer.io/fame/erogol/erogol/TTS/links/7)
## TTS Performance ## TTS Performance
<p align="center"><img src="https://camo.githubusercontent.com/9fa79f977015e55eb9ec7aa32045555f60d093d3/68747470733a2f2f646973636f757273652d706161732d70726f64756374696f6e2d636f6e74656e742e73332e6475616c737461636b2e75732d656173742d312e616d617a6f6e6177732e636f6d2f6f7074696d697a65642f33582f362f342f363432386639383065396563373531633234386535393134363038393566373838316165633063365f325f363930783339342e706e67"/></p> <p align="center"><img src="https://camo.githubusercontent.com/9fa79f977015e55eb9ec7aa32045555f60d093d3/68747470733a2f2f646973636f757273652d706161732d70726f64756374696f6e2d636f6e74656e742e73332e6475616c737461636b2e75732d656173742d312e616d617a6f6e6177732e636f6d2f6f7074696d697a65642f33582f362f342f363432386639383065396563373531633234386535393134363038393566373838316165633063365f325f363930783339342e706e67"/></p>
[Details...](https://github.com/mozilla/TTS/wiki/Mean-Opinion-Score-Results) [Details...](https://github.com/mozilla/TTS/wiki/Mean-Opinion-Score-Results)
## Provided Models and Methods
Text-to-Spectrogram:
- Tacotron: [paper](https://arxiv.org/abs/1703.10135)
- Tacotron2: [paper](https://arxiv.org/abs/1712.05884)
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/1907.09006)
- Double Decoder Consistency: [blog](https://erogol.com/solving-attention-problems-of-tts-models-with-double-decoder-consistency/)
Speaker Encoder:
- GE2E: [paper](https://arxiv.org/abs/1710.10467)
Vocoders:
- MelGAN: [paper](https://arxiv.org/abs/1710.10467)
- MultiBandMelGAN: [paper](https://arxiv.org/abs/2005.05106)
- GAN-TTS discriminators: [paper](https://arxiv.org/abs/1909.11646)
You can also help us implement more models. Some TTS related work can be found [here](https://github.com/erogol/TTS-papers).
## Features ## Features
- High performance Text2Speech models on Torch and Tensorflow 2.0. - High performance Deep Learning models for Text2Speech related tasks.
- High performance Speaker Encoder to compute speaker embeddings efficiently. - Text2Speech models (Tacotron, Tacotron2).
- Integration with various Neural Vocoders (PWGAN, MelGAN, WaveRNN) - Speaker Encoder to compute speaker embeddings efficiently.
- Released trained models. - Vocoder models (MelGAN, Multiband-MelGAN, GAN-TTS)
- Efficient training codes for PyTorch. (soon for Tensorflow 2.0) - Support for multi-speaker TTS training.
- Codes to convert Torch models to Tensorflow 2.0. - Support for Multi-GPUs training.
- Detailed training anlaysis on console and Tensorboard. - Ability to convert Torch models to Tensorflow 2.0 for inference.
- Released pre-trained models.
- Fast and efficient model training.
- Detailed training logs on console and Tensorboard.
- Tools to curate Text2Speech datasets under```dataset_analysis```. - Tools to curate Text2Speech datasets under```dataset_analysis```.
- Demo server for model testing. - Demo server for model testing.
- Notebooks for extensive model benchmarking. - Notebooks for extensive model benchmarking.
@ -45,6 +69,22 @@ Or you can use ```requirements.txt``` to install the requirements only.
```pip install -r requirements.txt``` ```pip install -r requirements.txt```
### Directory Structure
```
|- TTS/
| |- train.py (train your TTS model.)
| |- distribute.py (train your TTS model using Multiple GPUs)
| |- config.json (TTS model configuration file)
| |- tf/ (Tensorflow 2 utilities and model implementations)
| |- layers/ (model layer definitions)
| |- models/ (model definitions)
| |- notebooks/ (Jupyter Notebooks for model evaluation and parameter selection)
| |- data_analysis/ (TTS Dataset analysis tools and notebooks.)
| |- utils/ (TTS utilities -io, visualization, data processing etc.-)
| |- speaker_encoder/ (Speaker Encoder implementation with the same folder structure.)
| |- vocoder/ (Vocoder implementations with the same folder structure.)
```
### Docker ### Docker
A barebone `Dockerfile` exists at the root of the project, which should let you quickly setup the environment. By default, it will start the server and let you query it. Make sure to use `nvidia-docker` to use your GPUs. Make sure you follow the instructions in the [`server README`](server/README.md) before you build your image so that the server can find the model within the image. A barebone `Dockerfile` exists at the root of the project, which should let you quickly setup the environment. By default, it will start the server and let you query it. Make sure to use `nvidia-docker` to use your GPUs. Make sure you follow the instructions in the [`server README`](server/README.md) before you build your image so that the server can find the model within the image.
@ -61,7 +101,7 @@ Below you see Tacotron model state after 16K iterations with batch-size 32 with
> "Recent research at Harvard has shown meditating for as little as 8 weeks can actually increase the grey matter in the parts of the brain responsible for emotional regulation and learning." > "Recent research at Harvard has shown meditating for as little as 8 weeks can actually increase the grey matter in the parts of the brain responsible for emotional regulation and learning."
Audio examples: [https://soundcloud.com/user-565970875](https://soundcloud.com/user-565970875) Audio examples: [soundcloud](https://soundcloud.com/user-565970875/pocket-article-wavernn-and-tacotron2)
<img src="images/example_model_output.png?raw=true" alt="example_output" width="400"/> <img src="images/example_model_output.png?raw=true" alt="example_output" width="400"/>
@ -82,7 +122,7 @@ Audio length is approximately 6 secs.
## Datasets and Data-Loading ## Datasets and Data-Loading
TTS provides a generic dataloder easy to use for new datasets. You need to write an preprocessor function to integrate your own dataset.Check ```datasets/preprocess.py``` to see some examples. After the function, you need to set ```dataset``` field in ```config.json```. Do not forget other data related fields too. TTS provides a generic dataloader easy to use for new datasets. You need to write an preprocessor function to integrate your own dataset.Check ```datasets/preprocess.py``` to see some examples. After the function, you need to set ```dataset``` field in ```config.json```. Do not forget other data related fields too.
Some of the open-sourced datasets that we successfully applied TTS, are linked below. Some of the open-sourced datasets that we successfully applied TTS, are linked below.

8
compute_statistics.py
View File

@ -8,7 +8,7 @@ import numpy as np
from tqdm import tqdm from tqdm import tqdm
from TTS.datasets.preprocess import load_meta_data from TTS.datasets.preprocess import load_meta_data
from TTS.utils.generic_utils import load_config from TTS.utils.io import load_config
from TTS.utils.audio import AudioProcessor from TTS.utils.audio import AudioProcessor
def main(): def main():
@ -63,6 +63,11 @@ def main():
stats['linear_mean'] = linear_mean stats['linear_mean'] = linear_mean
stats['linear_std'] = linear_scale stats['linear_std'] = linear_scale
print(f' > Avg mel spec mean: {mel_mean.mean()}')
print(f' > Avg mel spec scale: {mel_scale.mean()}')
print(f' > Avg linear spec mean: {linear_mean.mean()}')
print(f' > Avg lienar spec scale: {linear_scale.mean()}')
# set default config values for mean-var scaling # set default config values for mean-var scaling
CONFIG.audio['stats_path'] = output_file_path CONFIG.audio['stats_path'] = output_file_path
CONFIG.audio['signal_norm'] = True CONFIG.audio['signal_norm'] = True
@ -73,6 +78,7 @@ def main():
del CONFIG.audio['clip_norm'] del CONFIG.audio['clip_norm']
stats['audio_config'] = CONFIG.audio stats['audio_config'] = CONFIG.audio
np.save(output_file_path, stats, allow_pickle=True) np.save(output_file_path, stats, allow_pickle=True)
print(f' > scale_stats.npy is saved to {output_file_path}')
if __name__ == "__main__": if __name__ == "__main__":

25
config.json
View File

@ -1,24 +1,24 @@
{ {
"model": "Tacotron2", "model": "Tacotron2",
"run_name": "ljspeech", "run_name": "ljspeech-ddc-bn",
"run_description": "tacotron2", "run_description": "tacotron2 with ddc and batch-normalization",
// AUDIO PARAMETERS // AUDIO PARAMETERS
"audio":{ "audio":{
// stft parameters // stft parameters
"num_freq": 513, // number of stft frequency levels. Size of the linear spectogram frame. "fft_size": 1024, // number of stft frequency levels. Size of the linear spectogram frame.
"win_length": 1024, // stft window length in ms. "win_length": 1024, // stft window length in ms.
"hop_length": 256, // stft window hop-lengh in ms. "hop_length": 256, // stft window hop-lengh in ms.
"frame_length_ms": null, // stft window length in ms.If null, 'win_length' is used. "frame_length_ms": null, // stft window length in ms.If null, 'win_length' is used.
"frame_shift_ms": null, // stft window hop-lengh in ms. If null, 'hop_length' is used. "frame_shift_ms": null, // stft window hop-lengh in ms. If null, 'hop_length' is used.
// Audio processing parameters // Audio processing parameters
"sample_rate": 22050, // DATASET-RELATED: wav sample-rate. If different than the original data, it is resampled. "sample_rate": 22050, // DATASET-RELATED: wav sample-rate.
"preemphasis": 0.0, // pre-emphasis to reduce spec noise and make it more structured. If 0.0, no -pre-emphasis. "preemphasis": 0.0, // pre-emphasis to reduce spec noise and make it more structured. If 0.0, no -pre-emphasis.
"ref_level_db": 20, // reference level db, theoretically 20db is the sound of air. "ref_level_db": 20, // reference level db, theoretically 20db is the sound of air.
// Silence trimming // Silence trimming
"do_trim_silence": true,// enable trimming of slience of audio as you load it. LJspeech (false), TWEB (false), Nancy (true) "do_trim_silence": true,// enable trimming of slience of audio as you load it. LJspeech (true), TWEB (false), Nancy (true)
"trim_db": 60, // threshold for timming silence. Set this according to your dataset. "trim_db": 60, // threshold for timming silence. Set this according to your dataset.
// Griffin-Lim // Griffin-Lim
@ -83,27 +83,30 @@
// TACOTRON PRENET // TACOTRON PRENET
"memory_size": -1, // ONLY TACOTRON - size of the memory queue used fro storing last decoder predictions for auto-regression. If < 0, memory queue is disabled and decoder only uses the last prediction frame. "memory_size": -1, // ONLY TACOTRON - size of the memory queue used fro storing last decoder predictions for auto-regression. If < 0, memory queue is disabled and decoder only uses the last prediction frame.
"prenet_type": "original", // "original" or "bn". "prenet_type": "bn", // "original" or "bn".
"prenet_dropout": true, // enable/disable dropout at prenet. "prenet_dropout": false, // enable/disable dropout at prenet.
// ATTENTION // ATTENTION
"attention_type": "original", // 'original' or 'graves' "attention_type": "original", // 'original' or 'graves'
"attention_heads": 4, // number of attention heads (only for 'graves') "attention_heads": 4, // number of attention heads (only for 'graves')
"attention_norm": "sigmoid", // softmax or sigmoid. Suggested to use softmax for Tacotron2 and sigmoid for Tacotron. "attention_norm": "sigmoid", // softmax or sigmoid.
"windowing": false, // Enables attention windowing. Used only in eval mode. "windowing": false, // Enables attention windowing. Used only in eval mode.
"use_forward_attn": false, // if it uses forward attention. In general, it aligns faster. "use_forward_attn": false, // if it uses forward attention. In general, it aligns faster.
"forward_attn_mask": false, // Additional masking forcing monotonicity only in eval mode. "forward_attn_mask": false, // Additional masking forcing monotonicity only in eval mode.
"transition_agent": false, // enable/disable transition agent of forward attention. "transition_agent": false, // enable/disable transition agent of forward attention.
"location_attn": true, // enable_disable location sensitive attention. It is enabled for TACOTRON by default. "location_attn": true, // enable_disable location sensitive attention. It is enabled for TACOTRON by default.
"bidirectional_decoder": false, // use https://arxiv.org/abs/1907.09006. Use it, if attention does not work well with your dataset. "bidirectional_decoder": false, // use https://arxiv.org/abs/1907.09006. Use it, if attention does not work well with your dataset.
"double_decoder_consistency": true, // use DDC explained here https://erogol.com/solving-attention-problems-of-tts-models-with-double-decoder-consistency-draft/
"ddc_r": 7, // reduction rate for coarse decoder.
// STOPNET // STOPNET
"stopnet": true, // Train stopnet predicting the end of synthesis. "stopnet": true, // Train stopnet predicting the end of synthesis.
"separate_stopnet": true, // Train stopnet seperately if 'stopnet==true'. It prevents stopnet loss to influence the rest of the model. It causes a better model, but it trains SLOWER. "separate_stopnet": true, // Train stopnet seperately if 'stopnet==true'. It prevents stopnet loss to influence the rest of the model. It causes a better model, but it trains SLOWER.
// TENSORBOARD and LOGGING // TENSORBOARD and LOGGING
"print_step": 25, // Number of steps to log traning on console. "print_step": 25, // Number of steps to log training on console.
"print_eval": false, // If True, it prints loss values in evalulation. "tb_plot_step:": 100, // Number of steps to plot TB training figures.
"print_eval": false, // If True, it prints intermediate loss values in evalulation.
"save_step": 10000, // Number of training steps expected to save traninpg stats and checkpoints. "save_step": 10000, // Number of training steps expected to save traninpg stats and checkpoints.
"checkpoint": true, // If true, it saves checkpoints per "save_step" "checkpoint": true, // If true, it saves checkpoints per "save_step"
"tb_model_param_stats": false, // true, plots param stats per layer on tensorboard. Might be memory consuming, but good for debugging. "tb_model_param_stats": false, // true, plots param stats per layer on tensorboard. Might be memory consuming, but good for debugging.
@ -122,7 +125,7 @@
// PHONEMES // PHONEMES
"phoneme_cache_path": "/media/erogol/data_ssd2/mozilla_us_phonemes_3", // phoneme computation is slow, therefore, it caches results in the given folder. "phoneme_cache_path": "/media/erogol/data_ssd2/mozilla_us_phonemes_3", // phoneme computation is slow, therefore, it caches results in the given folder.
"use_phonemes": false, // use phonemes instead of raw characters. It is suggested for better pronounciation. "use_phonemes": true, // use phonemes instead of raw characters. It is suggested for better pronounciation.
"phoneme_language": "en-us", // depending on your target language, pick one from https://github.com/bootphon/phonemizer#languages "phoneme_language": "en-us", // depending on your target language, pick one from https://github.com/bootphon/phonemizer#languages
// MULTI-SPEAKER and GST // MULTI-SPEAKER and GST

134
config_template.json
View File

@ -1,134 +0,0 @@
{
"model": "Tacotron2", // one of the model in models/
"run_name": "ljspeech-stft_params",
"run_description": "tacotron2 cosntant stf parameters",
// AUDIO PARAMETERS
"audio":{
// Audio processing parameters
"num_mels": 80, // size of the mel spec frame.
"num_freq": 513, // number of stft frequency levels. Size of the linear spectogram frame.
"sample_rate": 22050, // DATASET-RELATED: wav sample-rate. If different than the original data, it is resampled.
"win_length": 1024, // stft window length in ms.
"hop_length": 256, // stft window hop-lengh in ms.
"preemphasis": 0.0, // pre-emphasis to reduce spec noise and make it more structured. If 0.0, no -pre-emphasis.
"frame_length_ms": null, // stft window length in ms.If null, 'win_length' is used.
"frame_shift_ms": null, // stft window hop-lengh in ms. If null, 'hop_length' is used.
"min_level_db": -100, // normalization range
"ref_level_db": 20, // reference level db, theoretically 20db is the sound of air.
"power": 1.5, // value to sharpen wav signals after GL algorithm.
"griffin_lim_iters": 60,// #griffin-lim iterations. 30-60 is a good range. Larger the value, slower the generation.
// Normalization parameters
"signal_norm": true, // normalize the spec values in range [0, 1]
"symmetric_norm": true, // move normalization to range [-1, 1]
"max_norm": 1.0, // scale normalization to range [-max_norm, max_norm] or [0, max_norm]
"clip_norm": true, // clip normalized values into the range.
"mel_fmin": 0.0, // minimum freq level for mel-spec. ~50 for male and ~95 for female voices. Tune for dataset!!
"mel_fmax": 8000.0, // maximum freq level for mel-spec. Tune for dataset!!
"do_trim_silence": true, // enable trimming of slience of audio as you load it. LJspeech (false), TWEB (false), Nancy (true)
"trim_db": 60 // threshold for timming silence. Set this according to your dataset.
},
// VOCABULARY PARAMETERS
// if custom character set is not defined,
// default set in symbols.py is used
"characters":{
"pad": "_",
"eos": "~",
"bos": "^",
"characters": "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz!'(),-.:;? ",
"punctuations":"!'(),-.:;? ",
"phonemes":"iyɨʉɯuɪʏʊeøɘəɵɤoɛœɜɞʌɔæɐaɶɑɒᵻʘɓǀɗǃʄǂɠǁʛpbtdʈɖcɟkɡqɢʔɴŋɲɳnɱmʙrʀⱱɾɽɸβfvθðszʃʒʂʐçʝxɣχʁħʕhɦɬɮʋɹɻjɰlɭʎʟˈˌːˑʍwɥʜʢʡɕʑɺɧɚ˞ɫ"
},
// DISTRIBUTED TRAINING
"distributed":{
"backend": "nccl",
"url": "tcp:\/\/localhost:54321"
},
"reinit_layers": [], // give a list of layer names to restore from the given checkpoint. If not defined, it reloads all heuristically matching layers.
// TRAINING
"batch_size": 32, // Batch size for training. Lower values than 32 might cause hard to learn attention. It is overwritten by 'gradual_training'.
"eval_batch_size":16,
"r": 7, // Number of decoder frames to predict per iteration. Set the initial values if gradual training is enabled.
"gradual_training": [[0, 7, 64], [1, 5, 64], [50000, 3, 32], [130000, 2, 32], [290000, 1, 32]], //set gradual training steps [first_step, r, batch_size]. If it is null, gradual training is disabled. For Tacotron, you might need to reduce the 'batch_size' as you proceeed.
"loss_masking": true, // enable / disable loss masking against the sequence padding.
"ga_alpha": 10.0, // weight for guided attention loss. If > 0, guided attention is enabled.
// VALIDATION
"run_eval": true,
"test_delay_epochs": 10, //Until attention is aligned, testing only wastes computation time.
"test_sentences_file": null, // set a file to load sentences to be used for testing. If it is null then we use default english sentences.
// OPTIMIZER
"noam_schedule": false, // use noam warmup and lr schedule.
"grad_clip": 1.0, // upper limit for gradients for clipping.
"epochs": 1000, // total number of epochs to train.
"lr": 0.0001, // Initial learning rate. If Noam decay is active, maximum learning rate.
"wd": 0.000001, // Weight decay weight.
"warmup_steps": 4000, // Noam decay steps to increase the learning rate from 0 to "lr"
"seq_len_norm": false, // Normalize eash sample loss with its length to alleviate imbalanced datasets. Use it if your dataset is small or has skewed distribution of sequence lengths.
// TACOTRON PRENET
"memory_size": -1, // ONLY TACOTRON - size of the memory queue used fro storing last decoder predictions for auto-regression. If < 0, memory queue is disabled and decoder only uses the last prediction frame.
"prenet_type": "original", // "original" or "bn".
"prenet_dropout": true, // enable/disable dropout at prenet.
// ATTENTION
"attention_type": "original", // 'original' or 'graves'
"attention_heads": 4, // number of attention heads (only for 'graves')
"attention_norm": "sigmoid", // softmax or sigmoid. Suggested to use softmax for Tacotron2 and sigmoid for Tacotron.
"windowing": false, // Enables attention windowing. Used only in eval mode.
"use_forward_attn": false, // if it uses forward attention. In general, it aligns faster.
"forward_attn_mask": false, // Additional masking forcing monotonicity only in eval mode.
"transition_agent": false, // enable/disable transition agent of forward attention.
"location_attn": true, // enable_disable location sensitive attention. It is enabled for TACOTRON by default.
"bidirectional_decoder": false, // use https://arxiv.org/abs/1907.09006. Use it, if attention does not work well with your dataset.
// STOPNET
"stopnet": true, // Train stopnet predicting the end of synthesis.
"separate_stopnet": true, // Train stopnet seperately if 'stopnet==true'. It prevents stopnet loss to influence the rest of the model. It causes a better model, but it trains SLOWER.
// TENSORBOARD and LOGGING
"print_step": 25, // Number of steps to log traning on console.
"save_step": 10000, // Number of training steps expected to save traninpg stats and checkpoints.
"checkpoint": true, // If true, it saves checkpoints per "save_step"
"tb_model_param_stats": false, // true, plots param stats per layer on tensorboard. Might be memory consuming, but good for debugging.
// DATA LOADING
"text_cleaner": "phoneme_cleaners",
"enable_eos_bos_chars": false, // enable/disable beginning of sentence and end of sentence chars.
"num_loader_workers": 4, // number of training data loader processes. Don't set it too big. 4-8 are good values.
"num_val_loader_workers": 4, // number of evaluation data loader processes.
"batch_group_size": 0, //Number of batches to shuffle after bucketing.
"min_seq_len": 6, // DATASET-RELATED: minimum text length to use in training
"max_seq_len": 153, // DATASET-RELATED: maximum text length
// PATHS
"output_path": "/data4/rw/home/Trainings/",
// PHONEMES
"phoneme_cache_path": "mozilla_us_phonemes_3", // phoneme computation is slow, therefore, it caches results in the given folder.
"use_phonemes": true, // use phonemes instead of raw characters. It is suggested for better pronounciation.
"phoneme_language": "en-us", // depending on your target language, pick one from https://github.com/bootphon/phonemizer#languages
// MULTI-SPEAKER and GST
"use_speaker_embedding": false, // use speaker embedding to enable multi-speaker learning.
"style_wav_for_test": null, // path to style wav file to be used in TacotronGST inference.
"use_gst": false, // TACOTRON ONLY: use global style tokens
// DATASETS
"datasets": // List of datasets. They all merged and they get different speaker_ids.
[
{
"name": "ljspeech",
"path": "/root/LJSpeech-1.1/",
"meta_file_train": "metadata.csv",
"meta_file_val": null
}
]
}

2
datasets/TTSDataset.py
View File

@ -92,7 +92,7 @@ class MyDataset(Dataset):
return phonemes return phonemes
def _load_or_generate_phoneme_sequence(self, wav_file, text): def _load_or_generate_phoneme_sequence(self, wav_file, text):
file_name = os.path.basename(wav_file).split('.')[0] file_name = os.path.splitext(os.path.basename(wav_file))[0]
cache_path = os.path.join(self.phoneme_cache_path, cache_path = os.path.join(self.phoneme_cache_path,
file_name + '_phoneme.npy') file_name + '_phoneme.npy')
try: try:

6
datasets/preprocess.py
View File

@ -120,7 +120,7 @@ def mailabs(root_path, meta_files=None):
text = cols[1].strip() text = cols[1].strip()
items.append([text, wav_file, speaker_name]) items.append([text, wav_file, speaker_name])
else: else:
raise RuntimeError("> File %s is not exist!"%(wav_file)) raise RuntimeError("> File %s does not exist!"%(wav_file))
return items return items
@ -185,7 +185,7 @@ def libri_tts(root_path, meta_files=None):
text = cols[1] text = cols[1]
items.append([text, wav_file, speaker_name]) items.append([text, wav_file, speaker_name])
for item in items: for item in items:
assert os.path.exists(item[1]), f" [!] wav file is not exist - {item[1]}" assert os.path.exists(item[1]), f" [!] wav files don't exist - {item[1]}"
return items return items
@ -203,5 +203,5 @@ def custom_turkish(root_path, meta_file):
continue continue
text = cols[1].strip() text = cols[1].strip()
items.append([text, wav_file, speaker_name]) items.append([text, wav_file, speaker_name])
print(f" [!] {len(skipped_files)} files skipped. They are not exist...") print(f" [!] {len(skipped_files)} files skipped. They don't exist...")
return items return items

24
layers/custom_layers.py
View File

@ -1,24 +0,0 @@
# coding: utf-8
# import torch
# from torch import nn
# class StopProjection(nn.Module):
# r""" Simple projection layer to predict the "stop token"
# Args:
# in_features (int): size of the input vector
# out_features (int or list): size of each output vector. aka number
# of predicted frames.
# """
# def __init__(self, in_features, out_features):
# super(StopProjection, self).__init__()
# self.linear = nn.Linear(in_features, out_features)
# self.dropout = nn.Dropout(0.5)
# self.sigmoid = nn.Sigmoid()
# def forward(self, inputs):
# out = self.dropout(inputs)
# out = self.linear(out)
# out = self.sigmoid(out)
# return out

11
layers/losses.py
View File

@ -184,7 +184,7 @@ class TacotronLoss(torch.nn.Module):
def forward(self, postnet_output, decoder_output, mel_input, linear_input, def forward(self, postnet_output, decoder_output, mel_input, linear_input,
stopnet_output, stopnet_target, output_lens, decoder_b_output, stopnet_output, stopnet_target, output_lens, decoder_b_output,
alignments, alignment_lens, input_lens): alignments, alignment_lens, alignments_backwards, input_lens):
return_dict = {} return_dict = {}
# decoder and postnet losses # decoder and postnet losses
@ -226,6 +226,15 @@ class TacotronLoss(torch.nn.Module):
return_dict['decoder_b_loss'] = decoder_b_loss return_dict['decoder_b_loss'] = decoder_b_loss
return_dict['decoder_c_loss'] = decoder_c_loss return_dict['decoder_c_loss'] = decoder_c_loss
# double decoder consistency loss (if enabled)
if self.config.double_decoder_consistency:
decoder_b_loss = self.criterion(decoder_b_output, mel_input, output_lens)
# decoder_c_loss = torch.nn.functional.l1_loss(decoder_b_output, decoder_output)
attention_c_loss = torch.nn.functional.l1_loss(alignments, alignments_backwards)
loss += decoder_b_loss + attention_c_loss
return_dict['decoder_coarse_loss'] = decoder_b_loss
return_dict['decoder_ddc_loss'] = attention_c_loss
# guided attention loss (if enabled) # guided attention loss (if enabled)
if self.config.ga_alpha > 0: if self.config.ga_alpha > 0:
ga_loss = self.criterion_ga(alignments, input_lens, alignment_lens) ga_loss = self.criterion_ga(alignments, input_lens, alignment_lens)

143
models/tacotron.py
View File

@ -1,23 +1,21 @@
# coding: utf-8 # coding: utf-8
import torch import torch
import copy
from torch import nn from torch import nn
from TTS.layers.tacotron import Encoder, Decoder, PostCBHG
from TTS.utils.generic_utils import sequence_mask
from TTS.layers.gst_layers import GST from TTS.layers.gst_layers import GST
from TTS.layers.tacotron import Decoder, Encoder, PostCBHG
from TTS.models.tacotron_abstract import TacotronAbstract
class Tacotron(nn.Module): class Tacotron(TacotronAbstract):
def __init__(self, def __init__(self,
num_chars, num_chars,
num_speakers, num_speakers,
r=5, r=5,
postnet_output_dim=1025, postnet_output_dim=1025,
decoder_output_dim=80, decoder_output_dim=80,
memory_size=5,
attn_type='original', attn_type='original',
attn_win=False, attn_win=False,
gst=False,
attn_norm="sigmoid", attn_norm="sigmoid",
prenet_type="original", prenet_type="original",
prenet_dropout=True, prenet_dropout=True,
@ -27,38 +25,41 @@ class Tacotron(nn.Module):
location_attn=True, location_attn=True,
attn_K=5, attn_K=5,
separate_stopnet=True, separate_stopnet=True,
bidirectional_decoder=False): bidirectional_decoder=False,
super(Tacotron, self).__init__() double_decoder_consistency=False,
self.r = r ddc_r=None,
self.decoder_output_dim = decoder_output_dim gst=False,
self.postnet_output_dim = postnet_output_dim memory_size=5):
self.gst = gst super(Tacotron,
self.num_speakers = num_speakers self).__init__(num_chars, num_speakers, r, postnet_output_dim,
self.bidirectional_decoder = bidirectional_decoder decoder_output_dim, attn_type, attn_win,
decoder_dim = 512 if num_speakers > 1 else 256
encoder_dim = 512 if num_speakers > 1 else 256
proj_speaker_dim = 80 if num_speakers > 1 else 0
# embedding layer
self.embedding = nn.Embedding(num_chars, 256, padding_idx=0)
self.embedding.weight.data.normal_(0, 0.3)
# boilerplate model
self.encoder = Encoder(encoder_dim)
self.decoder = Decoder(decoder_dim, decoder_output_dim, r, memory_size, attn_type, attn_win,
attn_norm, prenet_type, prenet_dropout, attn_norm, prenet_type, prenet_dropout,
forward_attn, trans_agent, forward_attn_mask, forward_attn, trans_agent, forward_attn_mask,
location_attn, attn_K, separate_stopnet, location_attn, attn_K, separate_stopnet,
proj_speaker_dim) bidirectional_decoder, double_decoder_consistency,
if self.bidirectional_decoder: ddc_r, gst)
self.decoder_backward = copy.deepcopy(self.decoder) decoder_in_features = 512 if num_speakers > 1 else 256
encoder_in_features = 512 if num_speakers > 1 else 256
speaker_embedding_dim = 256
proj_speaker_dim = 80 if num_speakers > 1 else 0
# base model layers
self.embedding = nn.Embedding(num_chars, 256, padding_idx=0)
self.embedding.weight.data.normal_(0, 0.3)
self.encoder = Encoder(encoder_in_features)
self.decoder = Decoder(decoder_in_features, decoder_output_dim, r,
memory_size, attn_type, attn_win, attn_norm,
prenet_type, prenet_dropout, forward_attn,
trans_agent, forward_attn_mask, location_attn,
attn_K, separate_stopnet, proj_speaker_dim)
self.postnet = PostCBHG(decoder_output_dim) self.postnet = PostCBHG(decoder_output_dim)
self.last_linear = nn.Linear(self.postnet.cbhg.gru_features * 2, self.last_linear = nn.Linear(self.postnet.cbhg.gru_features * 2,
postnet_output_dim) postnet_output_dim)
# speaker embedding layers # speaker embedding layers
if num_speakers > 1: if num_speakers > 1:
self.speaker_embedding = nn.Embedding(num_speakers, 256) self.speaker_embedding = nn.Embedding(num_speakers, speaker_embedding_dim)
self.speaker_embedding.weight.data.normal_(0, 0.3) self.speaker_embedding.weight.data.normal_(0, 0.3)
self.speaker_project_mel = nn.Sequential( self.speaker_project_mel = nn.Sequential(
nn.Linear(256, proj_speaker_dim), nn.Tanh()) nn.Linear(speaker_embedding_dim, proj_speaker_dim), nn.Tanh())
self.speaker_embeddings = None self.speaker_embeddings = None
self.speaker_embeddings_projected = None self.speaker_embeddings_projected = None
# global style token layers # global style token layers
@ -68,28 +69,19 @@ class Tacotron(nn.Module):
num_heads=4, num_heads=4,
num_style_tokens=10, num_style_tokens=10,
embedding_dim=gst_embedding_dim) embedding_dim=gst_embedding_dim)
# backward pass decoder
if self.bidirectional_decoder:
self._init_backward_decoder()
# setup DDC
if self.double_decoder_consistency:
self.coarse_decoder = Decoder(
decoder_in_features, decoder_output_dim, ddc_r, memory_size,
attn_type, attn_win, attn_norm, prenet_type, prenet_dropout,
forward_attn, trans_agent, forward_attn_mask, location_attn,
attn_K, separate_stopnet, proj_speaker_dim)
def _init_states(self):
self.speaker_embeddings = None
self.speaker_embeddings_projected = None
def compute_speaker_embedding(self, speaker_ids): def forward(self, characters, text_lengths, mel_specs, mel_lengths=None, speaker_ids=None):
if hasattr(self, "speaker_embedding") and speaker_ids is None:
raise RuntimeError(
" [!] Model has speaker embedding layer but speaker_id is not provided"
)
if hasattr(self, "speaker_embedding") and speaker_ids is not None:
self.speaker_embeddings = self._compute_speaker_embedding(
speaker_ids)
self.speaker_embeddings_projected = self.speaker_project_mel(
self.speaker_embeddings).squeeze(1)
def compute_gst(self, inputs, mel_specs):
gst_outputs = self.gst_layer(mel_specs)
inputs = self._add_speaker_embedding(inputs, gst_outputs)
return inputs
def forward(self, characters, text_lengths, mel_specs, speaker_ids=None):
""" """
Shapes: Shapes:
- characters: B x T_in - characters: B x T_in
@ -98,37 +90,50 @@ class Tacotron(nn.Module):
- speaker_ids: B x 1 - speaker_ids: B x 1
""" """
self._init_states() self._init_states()
mask = sequence_mask(text_lengths).to(characters.device) input_mask, output_mask = self.compute_masks(text_lengths, mel_lengths)
# B x T_in x embed_dim # B x T_in x embed_dim
inputs = self.embedding(characters) inputs = self.embedding(characters)
# B x speaker_embed_dim # B x speaker_embed_dim
if speaker_ids is not None:
self.compute_speaker_embedding(speaker_ids) self.compute_speaker_embedding(speaker_ids)
if self.num_speakers > 1: if self.num_speakers > 1:
# B x T_in x embed_dim + speaker_embed_dim # B x T_in x embed_dim + speaker_embed_dim
inputs = self._concat_speaker_embedding(inputs, inputs = self._concat_speaker_embedding(inputs,
self.speaker_embeddings) self.speaker_embeddings)
# B x T_in x encoder_dim # B x T_in x encoder_in_features
encoder_outputs = self.encoder(inputs) encoder_outputs = self.encoder(inputs)
# sequence masking
encoder_outputs = encoder_outputs * input_mask.unsqueeze(2).expand_as(encoder_outputs)
# global style token
if self.gst: if self.gst:
# B x gst_dim # B x gst_dim
encoder_outputs = self.compute_gst(encoder_outputs, mel_specs) encoder_outputs = self.compute_gst(encoder_outputs, mel_specs)
if self.num_speakers > 1: if self.num_speakers > 1:
encoder_outputs = self._concat_speaker_embedding( encoder_outputs = self._concat_speaker_embedding(
encoder_outputs, self.speaker_embeddings) encoder_outputs, self.speaker_embeddings)
# decoder_outputs: B x decoder_dim x T_out # decoder_outputs: B x decoder_in_features x T_out
# alignments: B x T_in x encoder_dim # alignments: B x T_in x encoder_in_features
# stop_tokens: B x T_in # stop_tokens: B x T_in
decoder_outputs, alignments, stop_tokens = self.decoder( decoder_outputs, alignments, stop_tokens = self.decoder(
encoder_outputs, mel_specs, mask, encoder_outputs, mel_specs, input_mask,
self.speaker_embeddings_projected) self.speaker_embeddings_projected)
# B x T_out x decoder_dim # sequence masking
if output_mask is not None:
decoder_outputs = decoder_outputs * output_mask.unsqueeze(1).expand_as(decoder_outputs)
# B x T_out x decoder_in_features
postnet_outputs = self.postnet(decoder_outputs) postnet_outputs = self.postnet(decoder_outputs)
# sequence masking
if output_mask is not None:
postnet_outputs = postnet_outputs * output_mask.unsqueeze(2).expand_as(postnet_outputs)
# B x T_out x posnet_dim # B x T_out x posnet_dim
postnet_outputs = self.last_linear(postnet_outputs) postnet_outputs = self.last_linear(postnet_outputs)
# B x T_out x decoder_dim # B x T_out x decoder_in_features
decoder_outputs = decoder_outputs.transpose(1, 2).contiguous() decoder_outputs = decoder_outputs.transpose(1, 2).contiguous()
if self.bidirectional_decoder: if self.bidirectional_decoder:
decoder_outputs_backward, alignments_backward = self._backward_inference(mel_specs, encoder_outputs, mask) decoder_outputs_backward, alignments_backward = self._backward_pass(mel_specs, encoder_outputs, input_mask)
return decoder_outputs, postnet_outputs, alignments, stop_tokens, decoder_outputs_backward, alignments_backward
if self.double_decoder_consistency:
decoder_outputs_backward, alignments_backward = self._coarse_decoder_pass(mel_specs, encoder_outputs, alignments, input_mask)
return decoder_outputs, postnet_outputs, alignments, stop_tokens, decoder_outputs_backward, alignments_backward return decoder_outputs, postnet_outputs, alignments, stop_tokens, decoder_outputs_backward, alignments_backward
return decoder_outputs, postnet_outputs, alignments, stop_tokens return decoder_outputs, postnet_outputs, alignments, stop_tokens
@ -136,6 +141,7 @@ class Tacotron(nn.Module):
def inference(self, characters, speaker_ids=None, style_mel=None): def inference(self, characters, speaker_ids=None, style_mel=None):
inputs = self.embedding(characters) inputs = self.embedding(characters)
self._init_states() self._init_states()
if speaker_ids is not None:
self.compute_speaker_embedding(speaker_ids) self.compute_speaker_embedding(speaker_ids)
if self.num_speakers > 1: if self.num_speakers > 1:
inputs = self._concat_speaker_embedding(inputs, inputs = self._concat_speaker_embedding(inputs,
@ -152,28 +158,3 @@ class Tacotron(nn.Module):
postnet_outputs = self.last_linear(postnet_outputs) postnet_outputs = self.last_linear(postnet_outputs)
decoder_outputs = decoder_outputs.transpose(1, 2) decoder_outputs = decoder_outputs.transpose(1, 2)
return decoder_outputs, postnet_outputs, alignments, stop_tokens return decoder_outputs, postnet_outputs, alignments, stop_tokens
def _backward_inference(self, mel_specs, encoder_outputs, mask):
decoder_outputs_b, alignments_b, _ = self.decoder_backward(
encoder_outputs, torch.flip(mel_specs, dims=(1,)), mask,
self.speaker_embeddings_projected)
decoder_outputs_b = decoder_outputs_b.transpose(1, 2).contiguous()
return decoder_outputs_b, alignments_b
def _compute_speaker_embedding(self, speaker_ids):
speaker_embeddings = self.speaker_embedding(speaker_ids)
return speaker_embeddings.unsqueeze_(1)
@staticmethod
def _add_speaker_embedding(outputs, speaker_embeddings):
speaker_embeddings_ = speaker_embeddings.expand(
outputs.size(0), outputs.size(1), -1)
outputs = outputs + speaker_embeddings_
return outputs
@staticmethod
def _concat_speaker_embedding(outputs, speaker_embeddings):
speaker_embeddings_ = speaker_embeddings.expand(
outputs.size(0), outputs.size(1), -1)
outputs = torch.cat([outputs, speaker_embeddings_], dim=-1)
return outputs

149
models/tacotron2.py
View File

@ -1,13 +1,15 @@
import copy
import torch
from math import sqrt from math import sqrt
import torch
from torch import nn from torch import nn
from TTS.layers.tacotron2 import Encoder, Decoder, Postnet
from TTS.utils.generic_utils import sequence_mask from TTS.layers.gst_layers import GST
from TTS.layers.tacotron2 import Decoder, Encoder, Postnet
from TTS.models.tacotron_abstract import TacotronAbstract
# TODO: match function arguments with tacotron # TODO: match function arguments with tacotron
class Tacotron2(nn.Module): class Tacotron2(TacotronAbstract):
def __init__(self, def __init__(self,
num_chars, num_chars,
num_speakers, num_speakers,
@ -25,37 +27,48 @@ class Tacotron2(nn.Module):
location_attn=True, location_attn=True,
attn_K=5, attn_K=5,
separate_stopnet=True, separate_stopnet=True,
bidirectional_decoder=False): bidirectional_decoder=False,
super(Tacotron2, self).__init__() double_decoder_consistency=False,
self.postnet_output_dim = postnet_output_dim ddc_r=None,
self.decoder_output_dim = decoder_output_dim gst=False):
self.r = r super(Tacotron2,
self.bidirectional_decoder = bidirectional_decoder self).__init__(num_chars, num_speakers, r, postnet_output_dim,
decoder_dim = 512 if num_speakers > 1 else 512 decoder_output_dim, attn_type, attn_win,
encoder_dim = 512 if num_speakers > 1 else 512 attn_norm, prenet_type, prenet_dropout,
forward_attn, trans_agent, forward_attn_mask,
location_attn, attn_K, separate_stopnet,
bidirectional_decoder, double_decoder_consistency,
ddc_r, gst)
decoder_in_features = 512 if num_speakers > 1 else 512
encoder_in_features = 512 if num_speakers > 1 else 512
proj_speaker_dim = 80 if num_speakers > 1 else 0 proj_speaker_dim = 80 if num_speakers > 1 else 0
# embedding layer # base layers
self.embedding = nn.Embedding(num_chars, 512, padding_idx=0) self.embedding = nn.Embedding(num_chars, 512, padding_idx=0)
std = sqrt(2.0 / (num_chars + 512))
val = sqrt(3.0) * std # uniform bounds for std
self.embedding.weight.data.uniform_(-val, val)
if num_speakers > 1: if num_speakers > 1:
self.speaker_embedding = nn.Embedding(num_speakers, 512) self.speaker_embedding = nn.Embedding(num_speakers, 512)
self.speaker_embedding.weight.data.normal_(0, 0.3) self.speaker_embedding.weight.data.normal_(0, 0.3)
self.speaker_embeddings = None self.encoder = Encoder(encoder_in_features)
self.speaker_embeddings_projected = None self.decoder = Decoder(decoder_in_features, self.decoder_output_dim, r, attn_type, attn_win,
self.encoder = Encoder(encoder_dim)
self.decoder = Decoder(decoder_dim, self.decoder_output_dim, r, attn_type, attn_win,
attn_norm, prenet_type, prenet_dropout, attn_norm, prenet_type, prenet_dropout,
forward_attn, trans_agent, forward_attn_mask, forward_attn, trans_agent, forward_attn_mask,
location_attn, attn_K, separate_stopnet, proj_speaker_dim) location_attn, attn_K, separate_stopnet, proj_speaker_dim)
if self.bidirectional_decoder:
self.decoder_backward = copy.deepcopy(self.decoder)
self.postnet = Postnet(self.postnet_output_dim) self.postnet = Postnet(self.postnet_output_dim)
# global style token layers
def _init_states(self): if self.gst:
self.speaker_embeddings = None gst_embedding_dim = encoder_in_features
self.speaker_embeddings_projected = None self.gst_layer = GST(num_mel=80,
num_heads=4,
num_style_tokens=10,
embedding_dim=gst_embedding_dim)
# backward pass decoder
if self.bidirectional_decoder:
self._init_backward_decoder()
# setup DDC
if self.double_decoder_consistency:
self.coarse_decoder = Decoder(decoder_in_features, self.decoder_output_dim, ddc_r, attn_type, attn_win,
attn_norm, prenet_type, prenet_dropout,
forward_attn, trans_agent, forward_attn_mask,
location_attn, attn_K, separate_stopnet, proj_speaker_dim)
@staticmethod @staticmethod
def shape_outputs(mel_outputs, mel_outputs_postnet, alignments): def shape_outputs(mel_outputs, mel_outputs_postnet, alignments):
@ -63,22 +76,49 @@ class Tacotron2(nn.Module):
mel_outputs_postnet = mel_outputs_postnet.transpose(1, 2) mel_outputs_postnet = mel_outputs_postnet.transpose(1, 2)
return mel_outputs, mel_outputs_postnet, alignments return mel_outputs, mel_outputs_postnet, alignments
def forward(self, text, text_lengths, mel_specs=None, speaker_ids=None): def forward(self, text, text_lengths, mel_specs=None, mel_lengths=None, speaker_ids=None):
self._init_states() self._init_states()
# compute mask for padding # compute mask for padding
mask = sequence_mask(text_lengths).to(text.device) # B x T_in_max (boolean)
input_mask, output_mask = self.compute_masks(text_lengths, mel_lengths)
# B x D_embed x T_in_max
embedded_inputs = self.embedding(text).transpose(1, 2) embedded_inputs = self.embedding(text).transpose(1, 2)
# B x T_in_max x D_en
encoder_outputs = self.encoder(embedded_inputs, text_lengths) encoder_outputs = self.encoder(embedded_inputs, text_lengths)
# adding speaker embeddding to encoder output
# TODO: multi-speaker
# B x speaker_embed_dim
if speaker_ids is not None:
self.compute_speaker_embedding(speaker_ids)
if self.num_speakers > 1:
# B x T_in x embed_dim + speaker_embed_dim
encoder_outputs = self._add_speaker_embedding(encoder_outputs, encoder_outputs = self._add_speaker_embedding(encoder_outputs,
speaker_ids) self.speaker_embeddings)
encoder_outputs = encoder_outputs * input_mask.unsqueeze(2).expand_as(encoder_outputs)
# global style token
if self.gst:
# B x gst_dim
encoder_outputs = self.compute_gst(encoder_outputs, mel_specs)
# B x mel_dim x T_out -- B x T_out//r x T_in -- B x T_out//r
decoder_outputs, alignments, stop_tokens = self.decoder( decoder_outputs, alignments, stop_tokens = self.decoder(
encoder_outputs, mel_specs, mask) encoder_outputs, mel_specs, input_mask)
# sequence masking
if mel_lengths is not None:
decoder_outputs = decoder_outputs * output_mask.unsqueeze(1).expand_as(decoder_outputs)
# B x mel_dim x T_out
postnet_outputs = self.postnet(decoder_outputs) postnet_outputs = self.postnet(decoder_outputs)
postnet_outputs = decoder_outputs + postnet_outputs postnet_outputs = decoder_outputs + postnet_outputs
# sequence masking
if output_mask is not None:
postnet_outputs = postnet_outputs * output_mask.unsqueeze(1).expand_as(postnet_outputs)
# B x T_out x mel_dim -- B x T_out x mel_dim -- B x T_out//r x T_in
decoder_outputs, postnet_outputs, alignments = self.shape_outputs( decoder_outputs, postnet_outputs, alignments = self.shape_outputs(
decoder_outputs, postnet_outputs, alignments) decoder_outputs, postnet_outputs, alignments)
if self.bidirectional_decoder: if self.bidirectional_decoder:
decoder_outputs_backward, alignments_backward = self._backward_inference(mel_specs, encoder_outputs, mask) decoder_outputs_backward, alignments_backward = self._backward_pass(mel_specs, encoder_outputs, input_mask)
return decoder_outputs, postnet_outputs, alignments, stop_tokens, decoder_outputs_backward, alignments_backward
if self.double_decoder_consistency:
decoder_outputs_backward, alignments_backward = self._coarse_decoder_pass(mel_specs, encoder_outputs, alignments, input_mask)
return decoder_outputs, postnet_outputs, alignments, stop_tokens, decoder_outputs_backward, alignments_backward return decoder_outputs, postnet_outputs, alignments, stop_tokens, decoder_outputs_backward, alignments_backward
return decoder_outputs, postnet_outputs, alignments, stop_tokens return decoder_outputs, postnet_outputs, alignments, stop_tokens
@ -86,15 +126,18 @@ class Tacotron2(nn.Module):
def inference(self, text, speaker_ids=None): def inference(self, text, speaker_ids=None):
embedded_inputs = self.embedding(text).transpose(1, 2) embedded_inputs = self.embedding(text).transpose(1, 2)
encoder_outputs = self.encoder.inference(embedded_inputs) encoder_outputs = self.encoder.inference(embedded_inputs)
if speaker_ids is not None:
self.compute_speaker_embedding(speaker_ids)
if self.num_speakers > 1:
encoder_outputs = self._add_speaker_embedding(encoder_outputs, encoder_outputs = self._add_speaker_embedding(encoder_outputs,
speaker_ids) self.speaker_embeddings)
mel_outputs, alignments, stop_tokens = self.decoder.inference( decoder_outputs, alignments, stop_tokens = self.decoder.inference(
encoder_outputs) encoder_outputs)
mel_outputs_postnet = self.postnet(mel_outputs) postnet_outputs = self.postnet(decoder_outputs)
mel_outputs_postnet = mel_outputs + mel_outputs_postnet postnet_outputs = decoder_outputs + postnet_outputs
mel_outputs, mel_outputs_postnet, alignments = self.shape_outputs( decoder_outputs, postnet_outputs, alignments = self.shape_outputs(
mel_outputs, mel_outputs_postnet, alignments) decoder_outputs, postnet_outputs, alignments)
return mel_outputs, mel_outputs_postnet, alignments, stop_tokens return decoder_outputs, postnet_outputs, alignments, stop_tokens
def inference_truncated(self, text, speaker_ids=None): def inference_truncated(self, text, speaker_ids=None):
""" """
@ -112,22 +155,16 @@ class Tacotron2(nn.Module):
mel_outputs, mel_outputs_postnet, alignments) mel_outputs, mel_outputs_postnet, alignments)
return mel_outputs, mel_outputs_postnet, alignments, stop_tokens return mel_outputs, mel_outputs_postnet, alignments, stop_tokens
def _backward_inference(self, mel_specs, encoder_outputs, mask):
decoder_outputs_b, alignments_b, _ = self.decoder_backward(
encoder_outputs, torch.flip(mel_specs, dims=(1,)), mask,
self.speaker_embeddings_projected)
decoder_outputs_b = decoder_outputs_b.transpose(1, 2)
return decoder_outputs_b, alignments_b
def _add_speaker_embedding(self, encoder_outputs, speaker_ids): def _speaker_embedding_pass(self, encoder_outputs, speaker_ids):
if hasattr(self, "speaker_embedding") and speaker_ids is None: # TODO: multi-speaker
raise RuntimeError(" [!] Model has speaker embedding layer but speaker_id is not provided") # if hasattr(self, "speaker_embedding") and speaker_ids is None:
if hasattr(self, "speaker_embedding") and speaker_ids is not None: # raise RuntimeError(" [!] Model has speaker embedding layer but speaker_id is not provided")
speaker_embeddings = self.speaker_embedding(speaker_ids) # if hasattr(self, "speaker_embedding") and speaker_ids is not None:
speaker_embeddings.unsqueeze_(1) # speaker_embeddings = speaker_embeddings.expand(encoder_outputs.size(0),
speaker_embeddings = speaker_embeddings.expand(encoder_outputs.size(0), # encoder_outputs.size(1),
encoder_outputs.size(1), # -1)
-1) # encoder_outputs = encoder_outputs + speaker_embeddings
encoder_outputs = encoder_outputs + speaker_embeddings # return encoder_outputs
return encoder_outputs pass

180
models/tacotron_abstract.py Normal file
View File

@ -0,0 +1,180 @@
import copy
from abc import ABC, abstractmethod
import torch
from torch import nn
from TTS.utils.generic_utils import sequence_mask
class TacotronAbstract(ABC, nn.Module):
def __init__(self,
num_chars,
num_speakers,
r,
postnet_output_dim=80,
decoder_output_dim=80,
attn_type='original',
attn_win=False,
attn_norm="softmax",
prenet_type="original",
prenet_dropout=True,
forward_attn=False,
trans_agent=False,
forward_attn_mask=False,
location_attn=True,
attn_K=5,
separate_stopnet=True,
bidirectional_decoder=False,
double_decoder_consistency=False,
ddc_r=None,
gst=False):
""" Abstract Tacotron class """
super().__init__()
self.num_chars = num_chars
self.r = r
self.decoder_output_dim = decoder_output_dim
self.postnet_output_dim = postnet_output_dim
self.gst = gst
self.num_speakers = num_speakers
self.bidirectional_decoder = bidirectional_decoder
self.double_decoder_consistency = double_decoder_consistency
self.ddc_r = ddc_r
self.attn_type = attn_type
self.attn_win = attn_win
self.attn_norm = attn_norm
self.prenet_type = prenet_type
self.prenet_dropout = prenet_dropout
self.forward_attn = forward_attn
self.trans_agent = trans_agent
self.forward_attn_mask = forward_attn_mask
self.location_attn = location_attn
self.attn_K = attn_K
self.separate_stopnet = separate_stopnet
# layers
self.embedding = None
self.encoder = None
self.decoder = None
self.postnet = None
# global style token
if self.gst:
self.gst_layer = None
# model states
self.speaker_embeddings = None
self.speaker_embeddings_projected = None
# additional layers
self.decoder_backward = None
self.coarse_decoder = None
#############################
# INIT FUNCTIONS
#############################
def _init_states(self):
self.speaker_embeddings = None
self.speaker_embeddings_projected = None
def _init_backward_decoder(self):
self.decoder_backward = copy.deepcopy(self.decoder)
def _init_coarse_decoder(self):
self.coarse_decoder = copy.deepcopy(self.decoder)
self.coarse_decoder.r_init = self.ddc_r
self.coarse_decoder.set_r(self.ddc_r)
#############################
# CORE FUNCTIONS
#############################
@abstractmethod
def forward(self):
pass
@abstractmethod
def inference(self):
pass
#############################
# COMMON COMPUTE FUNCTIONS
#############################
def compute_masks(self, text_lengths, mel_lengths):
"""Compute masks against sequence paddings."""
# B x T_in_max (boolean)
device = text_lengths.device
input_mask = sequence_mask(text_lengths).to(device)
output_mask = None
if mel_lengths is not None:
max_len = mel_lengths.max()
r = self.decoder.r
max_len = max_len + (r - (max_len % r)) if max_len % r > 0 else max_len
output_mask = sequence_mask(mel_lengths, max_len=max_len).to(device)
return input_mask, output_mask
def _backward_pass(self, mel_specs, encoder_outputs, mask):
""" Run backwards decoder """
decoder_outputs_b, alignments_b, _ = self.decoder_backward(
encoder_outputs, torch.flip(mel_specs, dims=(1,)), mask,
self.speaker_embeddings_projected)
decoder_outputs_b = decoder_outputs_b.transpose(1, 2).contiguous()
return decoder_outputs_b, alignments_b
def _coarse_decoder_pass(self, mel_specs, encoder_outputs, alignments,
input_mask):
""" Double Decoder Consistency """
T = mel_specs.shape[1]
if T % self.coarse_decoder.r > 0:
padding_size = self.coarse_decoder.r - (T % self.coarse_decoder.r)
mel_specs = torch.nn.functional.pad(mel_specs,
(0, 0, 0, padding_size, 0, 0))
decoder_outputs_backward, alignments_backward, _ = self.coarse_decoder(
encoder_outputs.detach(), mel_specs, input_mask)
# scale_factor = self.decoder.r_init / self.decoder.r
alignments_backward = torch.nn.functional.interpolate(
alignments_backward.transpose(1, 2),
size=alignments.shape[1],
mode='nearest').transpose(1, 2)
decoder_outputs_backward = decoder_outputs_backward.transpose(1, 2)
decoder_outputs_backward = decoder_outputs_backward[:, :T, :]
return decoder_outputs_backward, alignments_backward
#############################
# EMBEDDING FUNCTIONS
#############################
def compute_speaker_embedding(self, speaker_ids):
""" Compute speaker embedding vectors """
if hasattr(self, "speaker_embedding") and speaker_ids is None:
raise RuntimeError(
" [!] Model has speaker embedding layer but speaker_id is not provided"
)
if hasattr(self, "speaker_embedding") and speaker_ids is not None:
self.speaker_embeddings = self.speaker_embedding(speaker_ids).unsqueeze(1)
if hasattr(self, "speaker_project_mel") and speaker_ids is not None:
self.speaker_embeddings_projected = self.speaker_project_mel(
self.speaker_embeddings).squeeze(1)
def compute_gst(self, inputs, mel_specs):
""" Compute global style token """
# pylint: disable=not-callable
gst_outputs = self.gst_layer(mel_specs)
inputs = self._add_speaker_embedding(inputs, gst_outputs)
return inputs
@staticmethod
def _add_speaker_embedding(outputs, speaker_embeddings):
speaker_embeddings_ = speaker_embeddings.expand(
outputs.size(0), outputs.size(1), -1)
outputs = outputs + speaker_embeddings_
return outputs
@staticmethod
def _concat_speaker_embedding(outputs, speaker_embeddings):
speaker_embeddings_ = speaker_embeddings.expand(
outputs.size(0), outputs.size(1), -1)
outputs = torch.cat([outputs, speaker_embeddings_], dim=-1)
return outputs

2
requirements.txt
View File

@ -1,5 +1,5 @@
numpy>=1.16.0 numpy>=1.16.0
torch>=0.4.1 torch>=1.5
librosa>=0.5.1 librosa>=0.5.1
Unidecode>=0.4.20 Unidecode>=0.4.20
tensorboard tensorboard

2
requirements_tests.txt
View File

@ -1,4 +1,5 @@
numpy>=1.16.0 numpy>=1.16.0
numba==0.48
torch>=0.4.1 torch>=0.4.1
tensorflow>=2.2 tensorflow>=2.2
librosa>=0.5.1 librosa>=0.5.1
@ -13,3 +14,4 @@ tqdm
soundfile soundfile
phonemizer phonemizer
bokeh==1.4.0 bokeh==1.4.0
nose

18
server/server.py
View File

@ -21,6 +21,8 @@ def create_argparser():
parser.add_argument('--pwgan_lib_path', type=str, default=None, help='path to ParallelWaveGAN project folder to be imported. If this is not passed, model uses Griffin-Lim for synthesis.') parser.add_argument('--pwgan_lib_path', type=str, default=None, help='path to ParallelWaveGAN project folder to be imported. If this is not passed, model uses Griffin-Lim for synthesis.')
parser.add_argument('--pwgan_file', type=str, default=None, help='path to ParallelWaveGAN checkpoint file.') parser.add_argument('--pwgan_file', type=str, default=None, help='path to ParallelWaveGAN checkpoint file.')
parser.add_argument('--pwgan_config', type=str, default=None, help='path to ParallelWaveGAN config file.') parser.add_argument('--pwgan_config', type=str, default=None, help='path to ParallelWaveGAN config file.')
parser.add_argument('--vocoder_config', type=str, default=None, help='path to TTS.vocoder config file.')
parser.add_argument('--vocoder_checkpoint', type=str, default=None, help='path to TTS.vocoder checkpoint file.')
parser.add_argument('--port', type=int, default=5002, help='port to listen on.') 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('--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('--debug', type=convert_boolean, default=False, help='true to enable Flask debug mode.')
@ -35,6 +37,11 @@ embedded_tts_folder = os.path.join(embedded_models_folder, 'tts')
tts_checkpoint_file = os.path.join(embedded_tts_folder, 'checkpoint.pth.tar') tts_checkpoint_file = os.path.join(embedded_tts_folder, 'checkpoint.pth.tar')
tts_config_file = os.path.join(embedded_tts_folder, 'config.json') tts_config_file = os.path.join(embedded_tts_folder, 'config.json')
embedded_vocoder_folder = os.path.join(embedded_models_folder, 'vocoder')
vocoder_checkpoint_file = os.path.join(embedded_vocoder_folder, 'checkpoint.pth.tar')
vocoder_config_file = os.path.join(embedded_vocoder_folder, 'config.json')
# These models are soon to be deprecated
embedded_wavernn_folder = os.path.join(embedded_models_folder, 'wavernn') embedded_wavernn_folder = os.path.join(embedded_models_folder, 'wavernn')
wavernn_checkpoint_file = os.path.join(embedded_wavernn_folder, 'checkpoint.pth.tar') wavernn_checkpoint_file = os.path.join(embedded_wavernn_folder, 'checkpoint.pth.tar')
wavernn_config_file = os.path.join(embedded_wavernn_folder, 'config.json') wavernn_config_file = os.path.join(embedded_wavernn_folder, 'config.json')
@ -50,6 +57,11 @@ if not args.tts_checkpoint and os.path.isfile(tts_checkpoint_file):
args.tts_checkpoint = tts_checkpoint_file args.tts_checkpoint = tts_checkpoint_file
if not args.tts_config and os.path.isfile(tts_config_file): if not args.tts_config and os.path.isfile(tts_config_file):
args.tts_config = tts_config_file args.tts_config = tts_config_file
if not args.vocoder_checkpoint and os.path.isfile(tts_checkpoint_file):
args.tts_checkpoint = tts_checkpoint_file
if not args.vocoder_config and os.path.isfile(tts_config_file):
args.tts_config = tts_config_file
if not args.wavernn_file and os.path.isfile(wavernn_checkpoint_file): if not args.wavernn_file and os.path.isfile(wavernn_checkpoint_file):
args.wavernn_file = wavernn_checkpoint_file args.wavernn_file = wavernn_checkpoint_file
if not args.wavernn_config and os.path.isfile(wavernn_config_file): if not args.wavernn_config and os.path.isfile(wavernn_config_file):
@ -76,5 +88,9 @@ def tts():
return send_file(data, mimetype='audio/wav') return send_file(data, mimetype='audio/wav')
if __name__ == '__main__': def main():
app.run(debug=args.debug, host='0.0.0.0', port=args.port) app.run(debug=args.debug, host='0.0.0.0', port=args.port)
if __name__ == '__main__':
main()

58
server/synthesizer.py
View File

@ -1,6 +1,7 @@
import io import io
import re import re
import sys import sys
import time
import numpy as np import numpy as np
import torch import torch
@ -10,6 +11,7 @@ from TTS.utils.audio import AudioProcessor
from TTS.utils.io import load_config from TTS.utils.io import load_config
from TTS.utils.generic_utils import setup_model from TTS.utils.generic_utils import setup_model
from TTS.utils.speakers import load_speaker_mapping from TTS.utils.speakers import load_speaker_mapping
from TTS.vocoder.utils.generic_utils import setup_generator
# pylint: disable=unused-wildcard-import # pylint: disable=unused-wildcard-import
# pylint: disable=wildcard-import # pylint: disable=wildcard-import
from TTS.utils.synthesis import * from TTS.utils.synthesis import *
@ -34,10 +36,12 @@ class Synthesizer(object):
assert torch.cuda.is_available(), "CUDA is not availabe on this machine." assert torch.cuda.is_available(), "CUDA is not availabe on this machine."
self.load_tts(self.config.tts_checkpoint, self.config.tts_config, self.load_tts(self.config.tts_checkpoint, self.config.tts_config,
self.config.use_cuda) self.config.use_cuda)
if self.config.vocoder_checkpoint:
self.load_vocoder(self.config.vocoder_checkpoint, self.config.vocoder_config, self.config.use_cuda)
if self.config.wavernn_lib_path: if self.config.wavernn_lib_path:
self.load_wavernn(self.config.wavernn_lib_path, self.config.wavernn_file, self.load_wavernn(self.config.wavernn_lib_path, self.config.wavernn_file,
self.config.wavernn_config, self.config.use_cuda) self.config.wavernn_config, self.config.use_cuda)
if self.config.pwgan_lib_path: if self.config.pwgan_file:
self.load_pwgan(self.config.pwgan_lib_path, self.config.pwgan_file, self.load_pwgan(self.config.pwgan_lib_path, self.config.pwgan_file,
self.config.pwgan_config, self.config.use_cuda) self.config.pwgan_config, self.config.use_cuda)
@ -77,6 +81,19 @@ class Synthesizer(object):
self.tts_model.decoder.max_decoder_steps = 3000 self.tts_model.decoder.max_decoder_steps = 3000
if 'r' in cp: if 'r' in cp:
self.tts_model.decoder.set_r(cp['r']) self.tts_model.decoder.set_r(cp['r'])
print(f" > model reduction factor: {cp['r']}")
def load_vocoder(self, model_file, model_config, use_cuda):
self.vocoder_config = load_config(model_config)
self.vocoder_model = setup_generator(self.vocoder_config)
self.vocoder_model.load_state_dict(torch.load(model_file, map_location="cpu")["model"])
self.vocoder_model.remove_weight_norm()
self.vocoder_model.inference_padding = 0
self.vocoder_config = load_config(model_config)
if use_cuda:
self.vocoder_model.cuda()
self.vocoder_model.eval()
def load_wavernn(self, lib_path, model_file, model_config, use_cuda): def load_wavernn(self, lib_path, model_file, model_config, use_cuda):
# TODO: set a function in wavernn code base for model setup and call it here. # TODO: set a function in wavernn code base for model setup and call it here.
@ -113,9 +130,14 @@ class Synthesizer(object):
self.wavernn.eval() self.wavernn.eval()
def load_pwgan(self, lib_path, model_file, model_config, use_cuda): def load_pwgan(self, lib_path, model_file, model_config, use_cuda):
sys.path.append(lib_path) # set this if ParallelWaveGAN is not installed globally if lib_path:
# set this if ParallelWaveGAN is not installed globally
sys.path.append(lib_path)
try:
#pylint: disable=import-outside-toplevel #pylint: disable=import-outside-toplevel
from parallel_wavegan.models import ParallelWaveGANGenerator from parallel_wavegan.models import ParallelWaveGANGenerator
except ImportError as e:
raise RuntimeError(f"cannot import parallel-wavegan, either install it or set its directory using the --pwgan_lib_path command line argument: {e}")
print(" > Loading PWGAN model ...") print(" > Loading PWGAN model ...")
print(" | > model config: ", model_config) print(" | > model config: ", model_config)
print(" | > model file: ", model_file) print(" | > model file: ", model_file)
@ -166,6 +188,7 @@ class Synthesizer(object):
return sentences return sentences
def tts(self, text, speaker_id=None): def tts(self, text, speaker_id=None):
start_time = time.time()
wavs = [] wavs = []
sens = self.split_into_sentences(text) sens = self.split_into_sentences(text)
print(sens) print(sens)
@ -179,24 +202,25 @@ class Synthesizer(object):
inputs = numpy_to_torch(inputs, torch.long, cuda=self.use_cuda) inputs = numpy_to_torch(inputs, torch.long, cuda=self.use_cuda)
inputs = inputs.unsqueeze(0) inputs = inputs.unsqueeze(0)
# synthesize voice # synthesize voice
decoder_output, postnet_output, alignments, stop_tokens = run_model_torch( decoder_output, postnet_output, alignments, stop_tokens = run_model_torch(self.tts_model, inputs, self.tts_config, False, speaker_id, None)
self.tts_model, inputs, self.tts_config, False, speaker_id, None)
# convert outputs to numpy # convert outputs to numpy
postnet_output, decoder_output, _, _ = parse_outputs_torch( if self.vocoder_model:
postnet_output, decoder_output, alignments, stop_tokens) vocoder_input = postnet_output[0].transpose(0, 1).unsqueeze(0)
wav = self.vocoder_model.inference(vocoder_input)
if self.pwgan:
vocoder_input = torch.FloatTensor(postnet_output.T).unsqueeze(0)
if self.use_cuda: if self.use_cuda:
vocoder_input.cuda() wav = wav.cpu().numpy()
wav = self.pwgan.inference(vocoder_input, hop_size=self.ap.hop_length) else:
wav = wav.numpy()
wav = wav.flatten()
elif self.wavernn: elif self.wavernn:
vocoder_input = torch.FloatTensor(postnet_output.T).unsqueeze(0) vocoder_input = None
if self.tts_config.model == "Tacotron":
vocoder_input = torch.FloatTensor(self.ap.out_linear_to_mel(linear_spec=postnet_output.T).T).T.unsqueeze(0)
else:
vocoder_input = postnet_output[0].transpose(0, 1).unsqueeze(0)
if self.use_cuda: if self.use_cuda:
vocoder_input.cuda() vocoder_input.cuda()
wav = self.wavernn.generate(vocoder_input, batched=self.config.is_wavernn_batched, target=11000, overlap=550) wav = self.wavernn.generate(vocoder_input, batched=self.config.is_wavernn_batched, target=11000, overlap=550)
else:
wav = inv_spectrogram(postnet_output, self.ap, self.tts_config)
# trim silence # trim silence
wav = trim_silence(wav, self.ap) wav = trim_silence(wav, self.ap)
@ -205,4 +229,10 @@ class Synthesizer(object):
out = io.BytesIO() out = io.BytesIO()
self.save_wav(wavs, out) self.save_wav(wavs, out)
# compute stats
process_time = time.time() - start_time
audio_time = len(wavs) / self.tts_config.audio['sample_rate']
print(f" > Processing time: {process_time}")
print(f" > Real-time factor: {process_time / audio_time}")
return out return out

2
server/templates/index.html
View File

@ -8,7 +8,7 @@
<meta name="description" content=""> <meta name="description" content="">
<meta name="author" content=""> <meta name="author" content="">
<title>Mozillia - Text2Speech engine</title> <title>Mozilla - Text2Speech engine</title>
<!-- Bootstrap core CSS --> <!-- Bootstrap core CSS -->
<link href="https://stackpath.bootstrapcdn.com/bootstrap/4.1.1/css/bootstrap.min.css" <link href="https://stackpath.bootstrapcdn.com/bootstrap/4.1.1/css/bootstrap.min.css"

11
setup.py
View File

@ -19,7 +19,7 @@ args, unknown_args = parser.parse_known_args()
# Remove our arguments from argv so that setuptools doesn't see them # Remove our arguments from argv so that setuptools doesn't see them
sys.argv = [sys.argv[0]] + unknown_args sys.argv = [sys.argv[0]] + unknown_args
version = '0.0.1' version = '0.0.3'
# Adapted from https://github.com/pytorch/pytorch # Adapted from https://github.com/pytorch/pytorch
cwd = os.path.dirname(os.path.abspath(__file__)) cwd = os.path.dirname(os.path.abspath(__file__))
@ -75,6 +75,11 @@ setup(
url='https://github.com/mozilla/TTS', url='https://github.com/mozilla/TTS',
description='Text to Speech with Deep Learning', description='Text to Speech with Deep Learning',
license='MPL-2.0', license='MPL-2.0',
entry_points={
'console_scripts': [
'tts-server = TTS.server.server:main'
]
},
package_dir={'': 'tts_namespace'}, package_dir={'': 'tts_namespace'},
packages=find_packages('tts_namespace'), packages=find_packages('tts_namespace'),
package_data={ package_data={
@ -92,8 +97,8 @@ setup(
}, },
install_requires=[ install_requires=[
"scipy>=0.19.0", "scipy>=0.19.0",
"torch>=0.4.1", "torch>=1.5",
"numpy==1.15.4", "numpy>=1.16.0",
"librosa==0.6.2", "librosa==0.6.2",
"unidecode==0.4.20", "unidecode==0.4.20",
"attrdict", "attrdict",

2
speaker_encoder/README.md
View File

@ -1,4 +1,4 @@
### Speaker embedding (Experimental) ### Speaker Encoder
This is an implementation of https://arxiv.org/abs/1710.10467. This model can be used for voice and speaker embedding. This is an implementation of https://arxiv.org/abs/1710.10467. This model can be used for voice and speaker embedding.

2
speaker_encoder/requirements.txt Normal file
View File

@ -0,0 +1,2 @@
umap-learn
numpy>=1.17.0

11
speaker_encoder/train.py
View File

@ -13,12 +13,11 @@ from TTS.speaker_encoder.model import SpeakerEncoder
from TTS.speaker_encoder.visual import plot_embeddings from TTS.speaker_encoder.visual import plot_embeddings
from TTS.speaker_encoder.generic_utils import save_best_model from TTS.speaker_encoder.generic_utils import save_best_model
from TTS.utils.audio import AudioProcessor from TTS.utils.audio import AudioProcessor
from TTS.utils.generic_utils import (NoamLR, check_update, copy_config_file, from TTS.utils.generic_utils import (create_experiment_folder, get_git_branch,
count_parameters,
create_experiment_folder, get_git_branch,
load_config,
remove_experiment_folder, set_init_dict) remove_experiment_folder, set_init_dict)
from TTS.utils.logger import Logger from TTS.utils.io import load_config, copy_config_file
from TTS.utils.training import check_update, NoamLR
from TTS.utils.tensorboard_logger import TensorboardLogger
from TTS.utils.radam import RAdam from TTS.utils.radam import RAdam
torch.backends.cudnn.enabled = True torch.backends.cudnn.enabled = True
@ -237,7 +236,7 @@ if __name__ == '__main__':
new_fields) new_fields)
LOG_DIR = OUT_PATH LOG_DIR = OUT_PATH
tb_logger = Logger(LOG_DIR) tb_logger = TensorboardLogger(LOG_DIR)
try: try:
main(args) main(args)

3
synthesize.py
View File

@ -7,7 +7,8 @@ import json
import string import string
from TTS.utils.synthesis import synthesis from TTS.utils.synthesis import synthesis
from TTS.utils.generic_utils import load_config, setup_model from TTS.utils.generic_utils import setup_model
from TTS.utils.io import load_config
from TTS.utils.text.symbols import make_symbols, symbols, phonemes from TTS.utils.text.symbols import make_symbols, symbols, phonemes
from TTS.utils.audio import AudioProcessor from TTS.utils.audio import AudioProcessor

BIN
tests/data/ljspeech/wavs/LJ001-0001.wav
View File

Binary file not shown.

2
tests/outputs/dummy_model_config.json
View File

@ -55,6 +55,8 @@
"separate_stopnet": true, // Train stopnet seperately if 'stopnet==true'. It prevents stopnet loss to influence the rest of the model. It causes a better model, but it trains SLOWER. "separate_stopnet": true, // Train stopnet seperately if 'stopnet==true'. It prevents stopnet loss to influence the rest of the model. It causes a better model, but it trains SLOWER.
"tb_model_param_stats": false, // true, plots param stats per layer on tensorboard. Might be memory consuming, but good for debugging. "tb_model_param_stats": false, // true, plots param stats per layer on tensorboard. Might be memory consuming, but good for debugging.
"use_gst": false, "use_gst": false,
"double_decoder_consistency": true, // use DDC explained here https://erogol.com/solving-attention-problems-of-tts-models-with-double-decoder-consistency-draft/
"ddc_r": 7, // reduction rate for coarse decoder.
"batch_size": 32, // Batch size for training. Lower values than 32 might cause hard to learn attention. "batch_size": 32, // Batch size for training. Lower values than 32 might cause hard to learn attention.
"eval_batch_size":16, "eval_batch_size":16,

2
tests/test_tacotron2_model.py
View File

@ -51,7 +51,7 @@ class TacotronTrainTest(unittest.TestCase):
optimizer = optim.Adam(model.parameters(), lr=c.lr) optimizer = optim.Adam(model.parameters(), lr=c.lr)
for i in range(5): for i in range(5):
mel_out, mel_postnet_out, align, stop_tokens = model.forward( mel_out, mel_postnet_out, align, stop_tokens = model.forward(
input, input_lengths, mel_spec, speaker_ids) input, input_lengths, mel_spec, mel_lengths, speaker_ids)
assert torch.sigmoid(stop_tokens).data.max() <= 1.0 assert torch.sigmoid(stop_tokens).data.max() <= 1.0
assert torch.sigmoid(stop_tokens).data.min() >= 0.0 assert torch.sigmoid(stop_tokens).data.min() >= 0.0
optimizer.zero_grad() optimizer.zero_grad()

5
tests/test_tacotron_model.py
View File

@ -66,7 +66,7 @@ class TacotronTrainTest(unittest.TestCase):
optimizer = optim.Adam(model.parameters(), lr=c.lr) optimizer = optim.Adam(model.parameters(), lr=c.lr)
for _ in range(5): for _ in range(5):
mel_out, linear_out, align, stop_tokens = model.forward( mel_out, linear_out, align, stop_tokens = model.forward(
input_dummy, input_lengths, mel_spec, speaker_ids) input_dummy, input_lengths, mel_spec, mel_lengths, speaker_ids)
optimizer.zero_grad() optimizer.zero_grad()
loss = criterion(mel_out, mel_spec, mel_lengths) loss = criterion(mel_out, mel_spec, mel_lengths)
stop_loss = criterion_st(stop_tokens, stop_targets) stop_loss = criterion_st(stop_tokens, stop_targets)
@ -95,6 +95,7 @@ class TacotronGSTTrainTest(unittest.TestCase):
mel_spec = torch.rand(8, 120, c.audio['num_mels']).to(device) mel_spec = torch.rand(8, 120, c.audio['num_mels']).to(device)
linear_spec = torch.rand(8, 120, c.audio['num_freq']).to(device) linear_spec = torch.rand(8, 120, c.audio['num_freq']).to(device)
mel_lengths = torch.randint(20, 120, (8, )).long().to(device) mel_lengths = torch.randint(20, 120, (8, )).long().to(device)
mel_lengths[-1] = 120
stop_targets = torch.zeros(8, 120, 1).float().to(device) stop_targets = torch.zeros(8, 120, 1).float().to(device)
speaker_ids = torch.randint(0, 5, (8, )).long().to(device) speaker_ids = torch.randint(0, 5, (8, )).long().to(device)
@ -130,7 +131,7 @@ class TacotronGSTTrainTest(unittest.TestCase):
optimizer = optim.Adam(model.parameters(), lr=c.lr) optimizer = optim.Adam(model.parameters(), lr=c.lr)
for _ in range(10): for _ in range(10):
mel_out, linear_out, align, stop_tokens = model.forward( mel_out, linear_out, align, stop_tokens = model.forward(
input_dummy, input_lengths, mel_spec, speaker_ids) input_dummy, input_lengths, mel_spec, mel_lengths, speaker_ids)
optimizer.zero_grad() optimizer.zero_grad()
loss = criterion(mel_out, mel_spec, mel_lengths) loss = criterion(mel_out, mel_spec, mel_lengths)
stop_loss = criterion_st(stop_tokens, stop_targets) stop_loss = criterion_st(stop_tokens, stop_targets)

177
train.py
View File

@ -20,7 +20,8 @@ from TTS.utils.generic_utils import (count_parameters, create_experiment_folder,
from TTS.utils.io import (save_best_model, save_checkpoint, from TTS.utils.io import (save_best_model, save_checkpoint,
load_config, copy_config_file) load_config, copy_config_file)
from TTS.utils.training import (NoamLR, check_update, adam_weight_decay, from TTS.utils.training import (NoamLR, check_update, adam_weight_decay,
gradual_training_scheduler, set_weight_decay) gradual_training_scheduler, set_weight_decay,
setup_torch_training_env)
from TTS.utils.tensorboard_logger import TensorboardLogger from TTS.utils.tensorboard_logger import TensorboardLogger
from TTS.utils.console_logger import ConsoleLogger from TTS.utils.console_logger import ConsoleLogger
from TTS.utils.speakers import load_speaker_mapping, save_speaker_mapping, \ from TTS.utils.speakers import load_speaker_mapping, save_speaker_mapping, \
@ -32,13 +33,8 @@ from TTS.datasets.preprocess import load_meta_data
from TTS.utils.radam import RAdam from TTS.utils.radam import RAdam
from TTS.utils.measures import alignment_diagonal_score from TTS.utils.measures import alignment_diagonal_score
torch.backends.cudnn.enabled = True
torch.backends.cudnn.benchmark = False use_cuda, num_gpus = setup_torch_training_env(True, False)
torch.manual_seed(54321)
use_cuda = torch.cuda.is_available()
num_gpus = torch.cuda.device_count()
print(" > Using CUDA: ", use_cuda)
print(" > Number of GPUs: ", num_gpus)
def setup_loader(ap, r, is_val=False, verbose=False): def setup_loader(ap, r, is_val=False, verbose=False):
@ -123,21 +119,7 @@ def train(model, criterion, optimizer, optimizer_st, scheduler,
verbose=(epoch == 0)) verbose=(epoch == 0))
model.train() model.train()
epoch_time = 0 epoch_time = 0
train_values = {
'avg_postnet_loss': 0,
'avg_decoder_loss': 0,
'avg_stopnet_loss': 0,
'avg_align_error': 0,
'avg_step_time': 0,
'avg_loader_time': 0
}
if c.bidirectional_decoder:
train_values['avg_decoder_b_loss'] = 0 # decoder backward loss
train_values['avg_decoder_c_loss'] = 0 # decoder consistency loss
if c.ga_alpha > 0:
train_values['avg_ga_loss'] = 0 # guidede attention loss
keep_avg = KeepAverage() keep_avg = KeepAverage()
keep_avg.add_values(train_values)
if use_cuda: if use_cuda:
batch_n_iter = int( batch_n_iter = int(
len(data_loader.dataset) / (c.batch_size * num_gpus)) len(data_loader.dataset) / (c.batch_size * num_gpus))
@ -162,13 +144,14 @@ def train(model, criterion, optimizer, optimizer_st, scheduler,
optimizer_st.zero_grad() optimizer_st.zero_grad()
# forward pass model # forward pass model
if c.bidirectional_decoder: if c.bidirectional_decoder or c.double_decoder_consistency:
decoder_output, postnet_output, alignments, stop_tokens, decoder_backward_output, alignments_backward = model( decoder_output, postnet_output, alignments, stop_tokens, decoder_backward_output, alignments_backward = model(
text_input, text_lengths, mel_input, speaker_ids=speaker_ids) text_input, text_lengths, mel_input, mel_lengths, speaker_ids=speaker_ids)
else: else:
decoder_output, postnet_output, alignments, stop_tokens = model( decoder_output, postnet_output, alignments, stop_tokens = model(
text_input, text_lengths, mel_input, speaker_ids=speaker_ids) text_input, text_lengths, mel_input, mel_lengths, speaker_ids=speaker_ids)
decoder_backward_output = None decoder_backward_output = None
alignments_backward = None
# set the alignment lengths wrt reduction factor for guided attention # set the alignment lengths wrt reduction factor for guided attention
if mel_lengths.max() % model.decoder.r != 0: if mel_lengths.max() % model.decoder.r != 0:
@ -180,12 +163,8 @@ def train(model, criterion, optimizer, optimizer_st, scheduler,
loss_dict = criterion(postnet_output, decoder_output, mel_input, loss_dict = criterion(postnet_output, decoder_output, mel_input,
linear_input, stop_tokens, stop_targets, linear_input, stop_tokens, stop_targets,
mel_lengths, decoder_backward_output, mel_lengths, decoder_backward_output,
alignments, alignment_lengths, text_lengths) alignments, alignment_lengths, alignments_backward,
if c.bidirectional_decoder: text_lengths)
keep_avg.update_values({'avg_decoder_b_loss': loss_dict['decoder_backward_loss'].item(),
'avg_decoder_c_loss': loss_dict['decoder_c_loss'].item()})
if c.ga_alpha > 0:
keep_avg.update_values({'avg_ga_loss': loss_dict['ga_loss'].item()})
# backward pass # backward pass
loss_dict['loss'].backward() loss_dict['loss'].backward()
@ -195,7 +174,6 @@ def train(model, criterion, optimizer, optimizer_st, scheduler,
# compute alignment error (the lower the better ) # compute alignment error (the lower the better )
align_error = 1 - alignment_diagonal_score(alignments) align_error = 1 - alignment_diagonal_score(alignments)
keep_avg.update_value('avg_align_error', align_error)
loss_dict['align_error'] = align_error loss_dict['align_error'] = align_error
# backpass and check the grad norm for stop loss # backpass and check the grad norm for stop loss
@ -210,23 +188,6 @@ def train(model, criterion, optimizer, optimizer_st, scheduler,
step_time = time.time() - start_time step_time = time.time() - start_time
epoch_time += step_time epoch_time += step_time
# update avg stats
update_train_values = {
'avg_postnet_loss': float(loss_dict['postnet_loss'].item()),
'avg_decoder_loss': float(loss_dict['decoder_loss'].item()),
'avg_stopnet_loss': loss_dict['stopnet_loss'].item() \
if isinstance(loss_dict['stopnet_loss'], float) else float(loss_dict['stopnet_loss'].item()),
'avg_step_time': step_time,
'avg_loader_time': loader_time
}
keep_avg.update_values(update_train_values)
if global_step % c.print_step == 0:
c_logger.print_train_step(batch_n_iter, num_iter, global_step,
avg_spec_length, avg_text_length,
step_time, loader_time, current_lr,
loss_dict, keep_avg.avg_values)
# aggregate losses from processes # aggregate losses from processes
if num_gpus > 1: if num_gpus > 1:
loss_dict['postnet_loss'] = reduce_tensor(loss_dict['postnet_loss'].data, num_gpus) loss_dict['postnet_loss'] = reduce_tensor(loss_dict['postnet_loss'].data, num_gpus)
@ -234,18 +195,41 @@ def train(model, criterion, optimizer, optimizer_st, scheduler,
loss_dict['loss'] = reduce_tensor(loss_dict['loss'] .data, num_gpus) loss_dict['loss'] = reduce_tensor(loss_dict['loss'] .data, num_gpus)
loss_dict['stopnet_loss'] = reduce_tensor(loss_dict['stopnet_loss'].data, num_gpus) if c.stopnet else loss_dict['stopnet_loss'] loss_dict['stopnet_loss'] = reduce_tensor(loss_dict['stopnet_loss'].data, num_gpus) if c.stopnet else loss_dict['stopnet_loss']
# detach loss values
loss_dict_new = dict()
for key, value in loss_dict.items():
if isinstance(value, (int, float)):
loss_dict_new[key] = value
else:
loss_dict_new[key] = value.item()
loss_dict = loss_dict_new
# update avg stats
update_train_values = dict()
for key, value in loss_dict.items():
update_train_values['avg_' + key] = value
update_train_values['avg_loader_time'] = loader_time
update_train_values['avg_step_time'] = step_time
keep_avg.update_values(update_train_values)
# print training progress
if global_step % c.print_step == 0:
c_logger.print_train_step(batch_n_iter, num_iter, global_step,
avg_spec_length, avg_text_length,
step_time, loader_time, current_lr,
loss_dict, keep_avg.avg_values)
if args.rank == 0: if args.rank == 0:
# Plot Training Iter Stats # Plot Training Iter Stats
# reduce TB load # reduce TB load
if global_step % 10 == 0: if global_step % c.tb_plot_step == 0:
iter_stats = { iter_stats = {
"loss_posnet": loss_dict['postnet_loss'].item(),
"loss_decoder": loss_dict['decoder_loss'].item(),
"lr": current_lr, "lr": current_lr,
"grad_norm": grad_norm, "grad_norm": grad_norm,
"grad_norm_st": grad_norm_st, "grad_norm_st": grad_norm_st,
"step_time": step_time "step_time": step_time
} }
iter_stats.update(loss_dict)
tb_logger.tb_train_iter_stats(global_step, iter_stats) tb_logger.tb_train_iter_stats(global_step, iter_stats)
if global_step % c.save_step == 0: if global_step % c.save_step == 0:
@ -253,7 +237,7 @@ def train(model, criterion, optimizer, optimizer_st, scheduler,
# save model # save model
save_checkpoint(model, optimizer, global_step, epoch, model.decoder.r, OUT_PATH, save_checkpoint(model, optimizer, global_step, epoch, model.decoder.r, OUT_PATH,
optimizer_st=optimizer_st, optimizer_st=optimizer_st,
model_loss=loss_dict['postnet_loss'].item()) model_loss=loss_dict['postnet_loss'])
# Diagnostic visualizations # Diagnostic visualizations
const_spec = postnet_output[0].data.cpu().numpy() const_spec = postnet_output[0].data.cpu().numpy()
@ -268,7 +252,7 @@ def train(model, criterion, optimizer, optimizer_st, scheduler,
"alignment": plot_alignment(align_img), "alignment": plot_alignment(align_img),
} }
if c.bidirectional_decoder: if c.bidirectional_decoder or c.double_decoder_consistency:
figures["alignment_backward"] = plot_alignment(alignments_backward[0].data.cpu().numpy()) figures["alignment_backward"] = plot_alignment(alignments_backward[0].data.cpu().numpy())
tb_logger.tb_train_figures(global_step, figures) tb_logger.tb_train_figures(global_step, figures)
@ -288,16 +272,8 @@ def train(model, criterion, optimizer, optimizer_st, scheduler,
# Plot Epoch Stats # Plot Epoch Stats
if args.rank == 0: if args.rank == 0:
# Plot Training Epoch Stats epoch_stats = {"epoch_time": epoch_time}
epoch_stats = { epoch_stats.update(keep_avg.avg_values)
"loss_postnet": keep_avg['avg_postnet_loss'],
"loss_decoder": keep_avg['avg_decoder_loss'],
"stopnet_loss": keep_avg['avg_stopnet_loss'],
"alignment_score": keep_avg['avg_align_error'],
"epoch_time": epoch_time
}
if c.ga_alpha > 0:
epoch_stats['guided_attention_loss'] = keep_avg['avg_ga_loss']
tb_logger.tb_train_epoch_stats(global_step, epoch_stats) tb_logger.tb_train_epoch_stats(global_step, epoch_stats)
if c.tb_model_param_stats: if c.tb_model_param_stats:
tb_logger.tb_model_weights(model, global_step) tb_logger.tb_model_weights(model, global_step)
@ -309,20 +285,7 @@ def evaluate(model, criterion, ap, global_step, epoch):
data_loader = setup_loader(ap, model.decoder.r, is_val=True) data_loader = setup_loader(ap, model.decoder.r, is_val=True)
model.eval() model.eval()
epoch_time = 0 epoch_time = 0
eval_values_dict = {
'avg_postnet_loss': 0,
'avg_decoder_loss': 0,
'avg_stopnet_loss': 0,
'avg_align_error': 0
}
if c.bidirectional_decoder:
eval_values_dict['avg_decoder_b_loss'] = 0 # decoder backward loss
eval_values_dict['avg_decoder_c_loss'] = 0 # decoder consistency loss
if c.ga_alpha > 0:
eval_values_dict['avg_ga_loss'] = 0 # guidede attention loss
keep_avg = KeepAverage() keep_avg = KeepAverage()
keep_avg.add_values(eval_values_dict)
c_logger.print_eval_start() c_logger.print_eval_start()
if data_loader is not None: if data_loader is not None:
for num_iter, data in enumerate(data_loader): for num_iter, data in enumerate(data_loader):
@ -333,13 +296,14 @@ def evaluate(model, criterion, ap, global_step, epoch):
assert mel_input.shape[1] % model.decoder.r == 0 assert mel_input.shape[1] % model.decoder.r == 0
# forward pass model # forward pass model
if c.bidirectional_decoder: if c.bidirectional_decoder or c.double_decoder_consistency:
decoder_output, postnet_output, alignments, stop_tokens, decoder_backward_output, alignments_backward = model( decoder_output, postnet_output, alignments, stop_tokens, decoder_backward_output, alignments_backward = model(
text_input, text_lengths, mel_input, speaker_ids=speaker_ids) text_input, text_lengths, mel_input, speaker_ids=speaker_ids)
else: else:
decoder_output, postnet_output, alignments, stop_tokens = model( decoder_output, postnet_output, alignments, stop_tokens = model(
text_input, text_lengths, mel_input, speaker_ids=speaker_ids) text_input, text_lengths, mel_input, speaker_ids=speaker_ids)
decoder_backward_output = None decoder_backward_output = None
alignments_backward = None
# set the alignment lengths wrt reduction factor for guided attention # set the alignment lengths wrt reduction factor for guided attention
if mel_lengths.max() % model.decoder.r != 0: if mel_lengths.max() % model.decoder.r != 0:
@ -351,12 +315,8 @@ def evaluate(model, criterion, ap, global_step, epoch):
loss_dict = criterion(postnet_output, decoder_output, mel_input, loss_dict = criterion(postnet_output, decoder_output, mel_input,
linear_input, stop_tokens, stop_targets, linear_input, stop_tokens, stop_targets,
mel_lengths, decoder_backward_output, mel_lengths, decoder_backward_output,
alignments, alignment_lengths, text_lengths) alignments, alignment_lengths, alignments_backward,
if c.bidirectional_decoder: text_lengths)
keep_avg.update_values({'avg_decoder_b_loss': loss_dict['decoder_b_loss'].item(),
'avg_decoder_c_loss': loss_dict['decoder_c_loss'].item()})
if c.ga_alpha > 0:
keep_avg.update_values({'avg_ga_loss': loss_dict['ga_loss'].item()})
# step time # step time
step_time = time.time() - start_time step_time = time.time() - start_time
@ -364,7 +324,7 @@ def evaluate(model, criterion, ap, global_step, epoch):
# compute alignment score # compute alignment score
align_error = 1 - alignment_diagonal_score(alignments) align_error = 1 - alignment_diagonal_score(alignments)
keep_avg.update_value('avg_align_error', align_error) loss_dict['align_error'] = align_error
# aggregate losses from processes # aggregate losses from processes
if num_gpus > 1: if num_gpus > 1:
@ -373,14 +333,20 @@ def evaluate(model, criterion, ap, global_step, epoch):
if c.stopnet: if c.stopnet:
loss_dict['stopnet_loss'] = reduce_tensor(loss_dict['stopnet_loss'].data, num_gpus) loss_dict['stopnet_loss'] = reduce_tensor(loss_dict['stopnet_loss'].data, num_gpus)
keep_avg.update_values({ # detach loss values
'avg_postnet_loss': loss_dict_new = dict()
float(loss_dict['postnet_loss'].item()), for key, value in loss_dict.items():
'avg_decoder_loss': if isinstance(value, (int, float)):
float(loss_dict['decoder_loss'].item()), loss_dict_new[key] = value
'avg_stopnet_loss': else:
float(loss_dict['stopnet_loss'].item()), loss_dict_new[key] = value.item()
}) loss_dict = loss_dict_new
# update avg stats
update_train_values = dict()
for key, value in loss_dict.items():
update_train_values['avg_' + key] = value
keep_avg.update_values(update_train_values)
if c.print_eval: if c.print_eval:
c_logger.print_eval_step(num_iter, loss_dict, keep_avg.avg_values) c_logger.print_eval_step(num_iter, loss_dict, keep_avg.avg_values)
@ -409,20 +375,11 @@ def evaluate(model, criterion, ap, global_step, epoch):
c.audio["sample_rate"]) c.audio["sample_rate"])
# Plot Validation Stats # Plot Validation Stats
epoch_stats = {
"loss_postnet": keep_avg['avg_postnet_loss'],
"loss_decoder": keep_avg['avg_decoder_loss'],
"stopnet_loss": keep_avg['avg_stopnet_loss'],
"alignment_score": keep_avg['avg_align_error'],
}
if c.bidirectional_decoder: if c.bidirectional_decoder or c.double_decoder_consistency:
epoch_stats['loss_decoder_backward'] = keep_avg['avg_decoder_b_loss']
align_b_img = alignments_backward[idx].data.cpu().numpy() align_b_img = alignments_backward[idx].data.cpu().numpy()
eval_figures['alignment_backward'] = plot_alignment(align_b_img) eval_figures['alignment2'] = plot_alignment(align_b_img)
if c.ga_alpha > 0: tb_logger.tb_eval_stats(global_step, keep_avg.avg_values)
epoch_stats['guided_attention_loss'] = keep_avg['avg_ga_loss']
tb_logger.tb_eval_stats(global_step, epoch_stats)
tb_logger.tb_eval_figures(global_step, eval_figures) tb_logger.tb_eval_figures(global_step, eval_figures)
if args.rank == 0 and epoch > c.test_delay_epochs: if args.rank == 0 and epoch > c.test_delay_epochs:
@ -431,7 +388,8 @@ def evaluate(model, criterion, ap, global_step, epoch):
"It took me quite a long time to develop a voice, and now that I have it I'm not going to be silent.", "It took me quite a long time to develop a voice, and now that I have it I'm not going to be silent.",
"Be a voice, not an echo.", "Be a voice, not an echo.",
"I'm sorry Dave. I'm afraid I can't do that.", "I'm sorry Dave. I'm afraid I can't do that.",
"This cake is great. It's so delicious and moist." "This cake is great. It's so delicious and moist.",
"Prior to November 22, 1963."
] ]
else: else:
with open(c.test_sentences_file, "r") as f: with open(c.test_sentences_file, "r") as f:
@ -516,8 +474,6 @@ def main(args): # pylint: disable=redefined-outer-name
model = setup_model(num_chars, num_speakers, c) model = setup_model(num_chars, num_speakers, c)
print(" | > Num output units : {}".format(ap.num_freq), flush=True)
params = set_weight_decay(model, c.wd) params = set_weight_decay(model, c.wd)
optimizer = RAdam(params, lr=c.lr, weight_decay=0) optimizer = RAdam(params, lr=c.lr, weight_decay=0)
if c.stopnet and c.separate_stopnet: if c.stopnet and c.separate_stopnet:
@ -542,7 +498,7 @@ def main(args): # pylint: disable=redefined-outer-name
except: except:
print(" > Partial model initialization.") print(" > Partial model initialization.")
model_dict = model.state_dict() model_dict = model.state_dict()
model_dict = set_init_dict(model_dict, checkpoint, c) model_dict = set_init_dict(model_dict, checkpoint['model'], c)
model.load_state_dict(model_dict) model.load_state_dict(model_dict)
del model_dict del model_dict
for group in optimizer.param_groups: for group in optimizer.param_groups:
@ -585,7 +541,6 @@ def main(args): # pylint: disable=redefined-outer-name
if c.bidirectional_decoder: if c.bidirectional_decoder:
model.decoder_backward.set_r(r) model.decoder_backward.set_r(r)
print("\n > Number of output frames:", model.decoder.r) print("\n > Number of output frames:", model.decoder.r)
train_avg_loss_dict, global_step = train(model, criterion, optimizer, train_avg_loss_dict, global_step = train(model, criterion, optimizer,
optimizer_st, scheduler, ap, optimizer_st, scheduler, ap,
global_step, epoch) global_step, epoch)
@ -667,7 +622,7 @@ if __name__ == '__main__':
os.chmod(OUT_PATH, 0o775) os.chmod(OUT_PATH, 0o775)
LOG_DIR = OUT_PATH LOG_DIR = OUT_PATH
tb_logger = TensorboardLogger(LOG_DIR) tb_logger = TensorboardLogger(LOG_DIR, model_name='TTS')
# write model desc to tensorboard # write model desc to tensorboard
tb_logger.tb_add_text('model-description', c['run_description'], 0) tb_logger.tb_add_text('model-description', c['run_description'], 0)

30
utils/audio.py
View File

@ -17,7 +17,7 @@ class AudioProcessor(object):
hop_length=None, hop_length=None,
win_length=None, win_length=None,
ref_level_db=None, ref_level_db=None,
num_freq=None, fft_size=1024,
power=None, power=None,
preemphasis=0.0, preemphasis=0.0,
signal_norm=None, signal_norm=None,
@ -25,6 +25,8 @@ class AudioProcessor(object):
max_norm=None, max_norm=None,
mel_fmin=None, mel_fmin=None,
mel_fmax=None, mel_fmax=None,
spec_gain=20,
stft_pad_mode='reflect',
clip_norm=True, clip_norm=True,
griffin_lim_iters=None, griffin_lim_iters=None,
do_trim_silence=False, do_trim_silence=False,
@ -41,7 +43,7 @@ class AudioProcessor(object):
self.frame_shift_ms = frame_shift_ms self.frame_shift_ms = frame_shift_ms
self.frame_length_ms = frame_length_ms self.frame_length_ms = frame_length_ms
self.ref_level_db = ref_level_db self.ref_level_db = ref_level_db
self.num_freq = num_freq self.fft_size = fft_size
self.power = power self.power = power
self.preemphasis = preemphasis self.preemphasis = preemphasis
self.griffin_lim_iters = griffin_lim_iters self.griffin_lim_iters = griffin_lim_iters
@ -49,6 +51,8 @@ class AudioProcessor(object):
self.symmetric_norm = symmetric_norm self.symmetric_norm = symmetric_norm
self.mel_fmin = mel_fmin or 0 self.mel_fmin = mel_fmin or 0
self.mel_fmax = mel_fmax self.mel_fmax = mel_fmax
self.spec_gain = float(spec_gain)
self.stft_pad_mode = 'reflect'
self.max_norm = 1.0 if max_norm is None else float(max_norm) self.max_norm = 1.0 if max_norm is None else float(max_norm)
self.clip_norm = clip_norm self.clip_norm = clip_norm
self.do_trim_silence = do_trim_silence self.do_trim_silence = do_trim_silence
@ -57,11 +61,12 @@ class AudioProcessor(object):
self.stats_path = stats_path self.stats_path = stats_path
# setup stft parameters # setup stft parameters
if hop_length is None: if hop_length is None:
self.n_fft, self.hop_length, self.win_length = self._stft_parameters() # compute stft parameters from given time values
self.hop_length, self.win_length = self._stft_parameters()
else: else:
# use stft parameters from config file
self.hop_length = hop_length self.hop_length = hop_length
self.win_length = win_length self.win_length = win_length
self.n_fft = (self.num_freq - 1) * 2
assert min_level_db != 0.0, " [!] min_level_db is 0" assert min_level_db != 0.0, " [!] min_level_db is 0"
members = vars(self) members = vars(self)
for key, value in members.items(): for key, value in members.items():
@ -84,19 +89,18 @@ class AudioProcessor(object):
assert self.mel_fmax <= self.sample_rate // 2 assert self.mel_fmax <= self.sample_rate // 2
return librosa.filters.mel( return librosa.filters.mel(
self.sample_rate, self.sample_rate,
self.n_fft, self.fft_size,
n_mels=self.num_mels, n_mels=self.num_mels,
fmin=self.mel_fmin, fmin=self.mel_fmin,
fmax=self.mel_fmax) fmax=self.mel_fmax)
def _stft_parameters(self, ): def _stft_parameters(self, ):
"""Compute necessary stft parameters with given time values""" """Compute necessary stft parameters with given time values"""
n_fft = (self.num_freq - 1) * 2
factor = self.frame_length_ms / self.frame_shift_ms factor = self.frame_length_ms / self.frame_shift_ms
assert (factor).is_integer(), " [!] frame_shift_ms should divide frame_length_ms" assert (factor).is_integer(), " [!] frame_shift_ms should divide frame_length_ms"
hop_length = int(self.frame_shift_ms / 1000.0 * self.sample_rate) hop_length = int(self.frame_shift_ms / 1000.0 * self.sample_rate)
win_length = int(hop_length * factor) win_length = int(hop_length * factor)
return n_fft, hop_length, win_length return hop_length, win_length
### normalization ### ### normalization ###
def _normalize(self, S): def _normalize(self, S):
@ -108,7 +112,7 @@ class AudioProcessor(object):
if hasattr(self, 'mel_scaler'): if hasattr(self, 'mel_scaler'):
if S.shape[0] == self.num_mels: if S.shape[0] == self.num_mels:
return self.mel_scaler.transform(S.T).T return self.mel_scaler.transform(S.T).T
elif S.shape[0] == self.n_fft / 2: elif S.shape[0] == self.fft_size / 2:
return self.linear_scaler.transform(S.T).T return self.linear_scaler.transform(S.T).T
else: else:
raise RuntimeError(' [!] Mean-Var stats does not match the given feature dimensions.') raise RuntimeError(' [!] Mean-Var stats does not match the given feature dimensions.')
@ -137,7 +141,7 @@ class AudioProcessor(object):
if hasattr(self, 'mel_scaler'): if hasattr(self, 'mel_scaler'):
if S_denorm.shape[0] == self.num_mels: if S_denorm.shape[0] == self.num_mels:
return self.mel_scaler.inverse_transform(S_denorm.T).T return self.mel_scaler.inverse_transform(S_denorm.T).T
elif S_denorm.shape[0] == self.n_fft / 2: elif S_denorm.shape[0] == self.fft_size / 2:
return self.linear_scaler.inverse_transform(S_denorm.T).T return self.linear_scaler.inverse_transform(S_denorm.T).T
else: else:
raise RuntimeError(' [!] Mean-Var stats does not match the given feature dimensions.') raise RuntimeError(' [!] Mean-Var stats does not match the given feature dimensions.')
@ -182,11 +186,11 @@ class AudioProcessor(object):
### DB and AMP conversion ### ### DB and AMP conversion ###
# pylint: disable=no-self-use # pylint: disable=no-self-use
def _amp_to_db(self, x): def _amp_to_db(self, x):
return 20 * np.log10(np.maximum(1e-5, x)) return self.spec_gain * np.log10(np.maximum(1e-5, x))
# pylint: disable=no-self-use # pylint: disable=no-self-use
def _db_to_amp(self, x): def _db_to_amp(self, x):
return np.power(10.0, x * 0.05) return np.power(10.0, x / self.spec_gain)
### Preemphasis ### ### Preemphasis ###
def apply_preemphasis(self, x): def apply_preemphasis(self, x):
@ -252,10 +256,10 @@ class AudioProcessor(object):
def _stft(self, y): def _stft(self, y):
return librosa.stft( return librosa.stft(
y=y, y=y,
n_fft=self.n_fft, n_fft=self.fft_size,
hop_length=self.hop_length, hop_length=self.hop_length,
win_length=self.win_length, win_length=self.win_length,
pad_mode='constant' pad_mode=self.stft_pad_mode,
) )
def _istft(self, y): def _istft(self, y):

2
utils/console_logger.py
View File

@ -87,7 +87,7 @@ class ConsoleLogger():
diff = 0 diff = 0
if self.old_eval_loss_dict is not None: if self.old_eval_loss_dict is not None:
diff = value - self.old_eval_loss_dict[key] diff = value - self.old_eval_loss_dict[key]
if diff < 0: if diff <= 0:
color = tcolors.OKGREEN color = tcolors.OKGREEN
sign = '' sign = ''
log_text += "{}{}:{} {:.5f} {}({}{:.5f})\n".format(indent, key, color, value, tcolors.ENDC, sign, diff) log_text += "{}{}:{} {:.5f} {}({}{:.5f})\n".format(indent, key, color, value, tcolors.ENDC, sign, diff)

29
utils/generic_utils.py
View File

@ -75,6 +75,7 @@ def split_dataset(items):
is_multi_speaker = len(set(speakers)) > 1 is_multi_speaker = len(set(speakers)) > 1
eval_split_size = 500 if len(items) * 0.01 > 500 else int( eval_split_size = 500 if len(items) * 0.01 > 500 else int(
len(items) * 0.01) len(items) * 0.01)
assert eval_split_size > 0, " [!] You do not have enough samples to train. You need at least 100 samples."
np.random.seed(0) np.random.seed(0)
np.random.shuffle(items) np.random.shuffle(items)
if is_multi_speaker: if is_multi_speaker:
@ -106,15 +107,15 @@ def sequence_mask(sequence_length, max_len=None):
return seq_range_expand < seq_length_expand return seq_range_expand < seq_length_expand
def set_init_dict(model_dict, checkpoint, c): def set_init_dict(model_dict, checkpoint_state, c):
# Partial initialization: if there is a mismatch with new and old layer, it is skipped. # Partial initialization: if there is a mismatch with new and old layer, it is skipped.
for k, v in checkpoint['model'].items(): for k, v in checkpoint_state.items():
if k not in model_dict: if k not in model_dict:
print(" | > Layer missing in the model definition: {}".format(k)) print(" | > Layer missing in the model definition: {}".format(k))
# 1. filter out unnecessary keys # 1. filter out unnecessary keys
pretrained_dict = { pretrained_dict = {
k: v k: v
for k, v in checkpoint['model'].items() if k in model_dict for k, v in checkpoint_state.items() if k in model_dict
} }
# 2. filter out different size layers # 2. filter out different size layers
pretrained_dict = { pretrained_dict = {
@ -145,7 +146,7 @@ def setup_model(num_chars, num_speakers, c):
model = MyModel(num_chars=num_chars, model = MyModel(num_chars=num_chars,
num_speakers=num_speakers, num_speakers=num_speakers,
r=c.r, r=c.r,
postnet_output_dim=c.audio['num_freq'], postnet_output_dim=int(c.audio['fft_size'] / 2 + 1),
decoder_output_dim=c.audio['num_mels'], decoder_output_dim=c.audio['num_mels'],
gst=c.use_gst, gst=c.use_gst,
memory_size=c.memory_size, memory_size=c.memory_size,
@ -160,13 +161,16 @@ def setup_model(num_chars, num_speakers, c):
location_attn=c.location_attn, location_attn=c.location_attn,
attn_K=c.attention_heads, attn_K=c.attention_heads,
separate_stopnet=c.separate_stopnet, separate_stopnet=c.separate_stopnet,
bidirectional_decoder=c.bidirectional_decoder) bidirectional_decoder=c.bidirectional_decoder,
double_decoder_consistency=c.double_decoder_consistency,
ddc_r=c.ddc_r)
elif c.model.lower() == "tacotron2": elif c.model.lower() == "tacotron2":
model = MyModel(num_chars=num_chars, model = MyModel(num_chars=num_chars,
num_speakers=num_speakers, num_speakers=num_speakers,
r=c.r, r=c.r,
postnet_output_dim=c.audio['num_mels'], postnet_output_dim=c.audio['num_mels'],
decoder_output_dim=c.audio['num_mels'], decoder_output_dim=c.audio['num_mels'],
gst=c.use_gst,
attn_type=c.attention_type, attn_type=c.attention_type,
attn_win=c.windowing, attn_win=c.windowing,
attn_norm=c.attention_norm, attn_norm=c.attention_norm,
@ -178,7 +182,9 @@ def setup_model(num_chars, num_speakers, c):
location_attn=c.location_attn, location_attn=c.location_attn,
attn_K=c.attention_heads, attn_K=c.attention_heads,
separate_stopnet=c.separate_stopnet, separate_stopnet=c.separate_stopnet,
bidirectional_decoder=c.bidirectional_decoder) bidirectional_decoder=c.bidirectional_decoder,
double_decoder_consistency=c.double_decoder_consistency,
ddc_r=c.ddc_r)
return model return model
class KeepAverage(): class KeepAverage():
@ -197,6 +203,11 @@ class KeepAverage():
self.iters[name] = init_iter self.iters[name] = init_iter
def update_value(self, name, value, weighted_avg=False): def update_value(self, name, value, weighted_avg=False):
if name not in self.avg_values:
# add value if not exist before
self.add_value(name, init_val=value)
else:
# else update existing value
if weighted_avg: if weighted_avg:
self.avg_values[name] = 0.99 * self.avg_values[name] + 0.01 * value self.avg_values[name] = 0.99 * self.avg_values[name] + 0.01 * value
self.iters[name] += 1 self.iters[name] += 1
@ -241,7 +252,7 @@ def check_config(c):
# audio processing parameters # audio processing parameters
_check_argument('num_mels', c['audio'], restricted=True, val_type=int, min_val=10, max_val=2056) _check_argument('num_mels', c['audio'], restricted=True, val_type=int, min_val=10, max_val=2056)
_check_argument('num_freq', c['audio'], restricted=True, val_type=int, min_val=128, max_val=4058) _check_argument('fft_size', c['audio'], restricted=True, val_type=int, min_val=128, max_val=4058)
_check_argument('sample_rate', c['audio'], restricted=True, val_type=int, min_val=512, max_val=100000) _check_argument('sample_rate', c['audio'], restricted=True, val_type=int, min_val=512, max_val=100000)
_check_argument('frame_length_ms', c['audio'], restricted=True, val_type=float, min_val=10, max_val=1000, alternative='win_length') _check_argument('frame_length_ms', c['audio'], restricted=True, val_type=float, min_val=10, max_val=1000, alternative='win_length')
_check_argument('frame_shift_ms', c['audio'], restricted=True, val_type=float, min_val=1, max_val=1000, alternative='hop_length') _check_argument('frame_shift_ms', c['audio'], restricted=True, val_type=float, min_val=1, max_val=1000, alternative='hop_length')
@ -267,6 +278,7 @@ def check_config(c):
_check_argument('clip_norm', c['audio'], restricted=True, val_type=bool) _check_argument('clip_norm', c['audio'], restricted=True, val_type=bool)
_check_argument('mel_fmin', c['audio'], restricted=True, val_type=float, min_val=0.0, max_val=1000) _check_argument('mel_fmin', c['audio'], restricted=True, val_type=float, min_val=0.0, max_val=1000)
_check_argument('mel_fmax', c['audio'], restricted=True, val_type=float, min_val=500.0) _check_argument('mel_fmax', c['audio'], restricted=True, val_type=float, min_val=500.0)
_check_argument('spec_gain', c['audio'], restricted=True, val_type=float, min_val=1, max_val=100)
_check_argument('do_trim_silence', c['audio'], restricted=True, val_type=bool) _check_argument('do_trim_silence', c['audio'], restricted=True, val_type=bool)
_check_argument('trim_db', c['audio'], restricted=True, val_type=int) _check_argument('trim_db', c['audio'], restricted=True, val_type=int)
@ -308,6 +320,8 @@ def check_config(c):
_check_argument('transition_agent', c, restricted=True, val_type=bool) _check_argument('transition_agent', c, restricted=True, val_type=bool)
_check_argument('location_attn', c, restricted=True, val_type=bool) _check_argument('location_attn', c, restricted=True, val_type=bool)
_check_argument('bidirectional_decoder', c, restricted=True, val_type=bool) _check_argument('bidirectional_decoder', c, restricted=True, val_type=bool)
_check_argument('double_decoder_consistency', c, restricted=True, val_type=bool)
_check_argument('ddc_r', c, restricted='double_decoder_consistency' in c.keys(), min_val=1, max_val=7, val_type=int)
# stopnet # stopnet
_check_argument('stopnet', c, restricted=True, val_type=bool) _check_argument('stopnet', c, restricted=True, val_type=bool)
@ -315,6 +329,7 @@ def check_config(c):
# tensorboard # tensorboard
_check_argument('print_step', c, restricted=True, val_type=int, min_val=1) _check_argument('print_step', c, restricted=True, val_type=int, min_val=1)
_check_argument('tb_plot_step', c, restricted=True, val_type=int, min_val=1)
_check_argument('save_step', c, restricted=True, val_type=int, min_val=1) _check_argument('save_step', c, restricted=True, val_type=int, min_val=1)
_check_argument('checkpoint', c, restricted=True, val_type=bool) _check_argument('checkpoint', c, restricted=True, val_type=bool)
_check_argument('tb_model_param_stats', c, restricted=True, val_type=bool) _check_argument('tb_model_param_stats', c, restricted=True, val_type=bool)

21
utils/tensorboard_logger.py
View File

@ -3,7 +3,8 @@ from tensorboardX import SummaryWriter
class TensorboardLogger(object): class TensorboardLogger(object):
def __init__(self, log_dir): def __init__(self, log_dir, model_name):
self.model_name = model_name
self.writer = SummaryWriter(log_dir) self.writer = SummaryWriter(log_dir)
self.train_stats = {} self.train_stats = {}
self.eval_stats = {} self.eval_stats = {}
@ -50,31 +51,31 @@ class TensorboardLogger(object):
traceback.print_exc() traceback.print_exc()
def tb_train_iter_stats(self, step, stats): def tb_train_iter_stats(self, step, stats):
self.dict_to_tb_scalar("TrainIterStats", stats, step) self.dict_to_tb_scalar(f"{self.model_name}_TrainIterStats", stats, step)
def tb_train_epoch_stats(self, step, stats): def tb_train_epoch_stats(self, step, stats):
self.dict_to_tb_scalar("TrainEpochStats", stats, step) self.dict_to_tb_scalar(f"{self.model_name}_TrainEpochStats", stats, step)
def tb_train_figures(self, step, figures): def tb_train_figures(self, step, figures):
self.dict_to_tb_figure("TrainFigures", figures, step) self.dict_to_tb_figure(f"{self.model_name}_TrainFigures", figures, step)
def tb_train_audios(self, step, audios, sample_rate): def tb_train_audios(self, step, audios, sample_rate):
self.dict_to_tb_audios("TrainAudios", audios, step, sample_rate) self.dict_to_tb_audios(f"{self.model_name}_TrainAudios", audios, step, sample_rate)
def tb_eval_stats(self, step, stats): def tb_eval_stats(self, step, stats):
self.dict_to_tb_scalar("EvalStats", stats, step) self.dict_to_tb_scalar(f"{self.model_name}_EvalStats", stats, step)
def tb_eval_figures(self, step, figures): def tb_eval_figures(self, step, figures):
self.dict_to_tb_figure("EvalFigures", figures, step) self.dict_to_tb_figure(f"{self.model_name}_EvalFigures", figures, step)
def tb_eval_audios(self, step, audios, sample_rate): def tb_eval_audios(self, step, audios, sample_rate):
self.dict_to_tb_audios("EvalAudios", audios, step, sample_rate) self.dict_to_tb_audios(f"{self.model_name}_EvalAudios", audios, step, sample_rate)
def tb_test_audios(self, step, audios, sample_rate): def tb_test_audios(self, step, audios, sample_rate):
self.dict_to_tb_audios("TestAudios", audios, step, sample_rate) self.dict_to_tb_audios(f"{self.model_name}_TestAudios", audios, step, sample_rate)
def tb_test_figures(self, step, figures): def tb_test_figures(self, step, figures):
self.dict_to_tb_figure("TestFigures", figures, step) self.dict_to_tb_figure(f"{self.model_name}_TestFigures", figures, step)
def tb_add_text(self, title, text, step): def tb_add_text(self, title, text, step):
self.writer.add_text(title, text, step) self.writer.add_text(title, text, step)

1
utils/text/__init__.py
View File

@ -77,7 +77,6 @@ def phoneme_to_sequence(text, cleaner_names, language, enable_eos_bos=False, tp=
_phonemes_to_id = {s: i for i, s in enumerate(_phonemes)} _phonemes_to_id = {s: i for i, s in enumerate(_phonemes)}
sequence = [] sequence = []
text = text.replace(":", "")
clean_text = _clean_text(text, cleaner_names) clean_text = _clean_text(text, cleaner_names)
to_phonemes = text2phone(clean_text, language) to_phonemes = text2phone(clean_text, language)
if to_phonemes is None: if to_phonemes is None:

2
utils/text/cleaners.py
View File

@ -71,7 +71,7 @@ def remove_aux_symbols(text):
def replace_symbols(text): def replace_symbols(text):
text = text.replace(';', ',') text = text.replace(';', ',')
text = text.replace('-', ' ') text = text.replace('-', ' ')
text = text.replace(':', ' ') text = text.replace(':', ',')
text = text.replace('&', 'and') text = text.replace('&', 'and')
return text return text

103
utils/text/number_norm.py
View File

@ -1,51 +1,18 @@
# -*- coding: utf-8 -*-
""" from https://github.com/keithito/tacotron """
import inflect
import re import re
_inflect = inflect.engine()
_comma_number_re = re.compile(r'([0-9][0-9\,]+[0-9])') _comma_number_re = re.compile(r'([0-9][0-9\,]+[0-9])')
_decimal_number_re = re.compile(r'([0-9]+\.[0-9]+)') _decimal_number_re = re.compile(r'([0-9]+\.[0-9]+)')
_pounds_re = re.compile(r'£([0-9\,]*[0-9]+)') _pounds_re = re.compile(r'£([0-9\,]*[0-9]+)')
_dollars_re = re.compile(r'\$([0-9\.\,]*[0-9]+)') _dollars_re = re.compile(r'\$([0-9\.\,]*[0-9]+)')
_ordinal_re = re.compile(r'([0-9]+)(st|nd|rd|th)') _ordinal_re = re.compile(r'[0-9]+(st|nd|rd|th)')
_number_re = re.compile(r'[0-9]+') _number_re = re.compile(r'[0-9]+')
_units = [
'', 'one', 'two', 'three', 'four', 'five', 'six', 'seven', 'eight', 'nine',
'ten', 'eleven', 'twelve', 'thirteen', 'fourteen', 'fifteen', 'sixteen',
'seventeen', 'eighteen', 'nineteen'
]
_tens = [
'',
'ten',
'twenty',
'thirty',
'forty',
'fifty',
'sixty',
'seventy',
'eighty',
'ninety',
]
_digit_groups = [
'',
'thousand',
'million',
'billion',
'trillion',
'quadrillion',
]
_ordinal_suffixes = [
('one', 'first'),
('two', 'second'),
('three', 'third'),
('five', 'fifth'),
('eight', 'eighth'),
('nine', 'ninth'),
('twelve', 'twelfth'),
('ty', 'tieth'),
]
def _remove_commas(m): def _remove_commas(m):
return m.group(1).replace(',', '') return m.group(1).replace(',', '')
@ -66,55 +33,33 @@ def _expand_dollars(m):
dollar_unit = 'dollar' if dollars == 1 else 'dollars' dollar_unit = 'dollar' if dollars == 1 else 'dollars'
cent_unit = 'cent' if cents == 1 else 'cents' cent_unit = 'cent' if cents == 1 else 'cents'
return '%s %s, %s %s' % (dollars, dollar_unit, cents, cent_unit) return '%s %s, %s %s' % (dollars, dollar_unit, cents, cent_unit)
if dollars: elif dollars:
dollar_unit = 'dollar' if dollars == 1 else 'dollars' dollar_unit = 'dollar' if dollars == 1 else 'dollars'
return '%s %s' % (dollars, dollar_unit) return '%s %s' % (dollars, dollar_unit)
if cents: elif cents:
cent_unit = 'cent' if cents == 1 else 'cents' cent_unit = 'cent' if cents == 1 else 'cents'
return '%s %s' % (cents, cent_unit) return '%s %s' % (cents, cent_unit)
else:
return 'zero dollars' return 'zero dollars'
def _standard_number_to_words(n, digit_group): def _expand_ordinal(m):
parts = [] return _inflect.number_to_words(m.group(0))
if n >= 1000:
# Format next higher digit group.
parts.append(_standard_number_to_words(n // 1000, digit_group + 1))
n = n % 1000
if n >= 100:
parts.append('%s hundred' % _units[n // 100])
if n % 100 >= len(_units):
parts.append(_tens[(n % 100) // 10])
parts.append(_units[(n % 100) % 10])
else:
parts.append(_units[n % 100])
if n > 0:
parts.append(_digit_groups[digit_group])
return ' '.join([x for x in parts if x])
def _number_to_words(n):
# Handle special cases first, then go to the standard case:
if n >= 1000000000000000000:
return str(n) # Too large, just return the digits
if n == 0:
return 'zero'
if n % 100 == 0 and n % 1000 != 0 and n < 3000:
return _standard_number_to_words(n // 100, 0) + ' hundred'
return _standard_number_to_words(n, 0)
def _expand_number(m): def _expand_number(m):
return _number_to_words(int(m.group(0))) num = int(m.group(0))
if num > 1000 and num < 3000:
if num == 2000:
def _expand_ordinal(m): return 'two thousand'
num = _number_to_words(int(m.group(1))) elif num > 2000 and num < 2010:
for suffix, replacement in _ordinal_suffixes: return 'two thousand ' + _inflect.number_to_words(num % 100)
if num.endswith(suffix): elif num % 100 == 0:
return num[:-len(suffix)] + replacement return _inflect.number_to_words(num // 100) + ' hundred'
return num + 'th' else:
return _inflect.number_to_words(num, andword='', zero='oh', group=2).replace(', ', ' ')
else:
return _inflect.number_to_words(num, andword='')
def normalize_numbers(text): def normalize_numbers(text):

17
utils/training.py
View File

@ -2,6 +2,17 @@ import torch
import numpy as np import numpy as np
def setup_torch_training_env(cudnn_enable, cudnn_benchmark):
torch.backends.cudnn.enabled = cudnn_enable
torch.backends.cudnn.benchmark = cudnn_benchmark
torch.manual_seed(54321)
use_cuda = torch.cuda.is_available()
num_gpus = torch.cuda.device_count()
print(" > Using CUDA: ", use_cuda)
print(" > Number of GPUs: ", num_gpus)
return use_cuda, num_gpus
def check_update(model, grad_clip, ignore_stopnet=False): def check_update(model, grad_clip, ignore_stopnet=False):
r'''Check model gradient against unexpected jumps and failures''' r'''Check model gradient against unexpected jumps and failures'''
skip_flag = False skip_flag = False
@ -9,6 +20,12 @@ def check_update(model, grad_clip, ignore_stopnet=False):
grad_norm = torch.nn.utils.clip_grad_norm_([param for name, param in model.named_parameters() if 'stopnet' not in name], grad_clip) grad_norm = torch.nn.utils.clip_grad_norm_([param for name, param in model.named_parameters() if 'stopnet' not in name], grad_clip)
else: else:
grad_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), grad_clip) grad_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), grad_clip)
# compatibility with different torch versions
if isinstance(grad_norm, float):
if np.isinf(grad_norm):
print(" | > Gradient is INF !!")
skip_flag = True
else:
if torch.isinf(grad_norm): if torch.isinf(grad_norm):
print(" | > Gradient is INF !!") print(" | > Gradient is INF !!")
skip_flag = True skip_flag = True

13
utils/visual.py
View File

@ -27,14 +27,15 @@ def plot_alignment(alignment, info=None, fig_size=(16, 10), title=None):
return fig return fig
def plot_spectrogram(linear_output, audio, fig_size=(16, 10)): def plot_spectrogram(spectrogram, ap=None, fig_size=(16, 10)):
if isinstance(linear_output, torch.Tensor): if isinstance(spectrogram, torch.Tensor):
linear_output_ = linear_output.detach().cpu().numpy().squeeze() spectrogram_ = spectrogram.detach().cpu().numpy().squeeze().T
else: else:
linear_output_ = linear_output spectrogram_ = spectrogram.T
spectrogram = audio._denormalize(linear_output_.T) # pylint: disable=protected-access if ap is not None:
spectrogram_ = ap._denormalize(spectrogram_) # pylint: disable=protected-access
fig = plt.figure(figsize=fig_size) fig = plt.figure(figsize=fig_size)
plt.imshow(spectrogram, aspect="auto", origin="lower") plt.imshow(spectrogram_, aspect="auto", origin="lower")
plt.colorbar() plt.colorbar()
plt.tight_layout() plt.tight_layout()
return fig return fig

38
vocoder/README.md Normal file
View File

@ -0,0 +1,38 @@
# Mozilla TTS Vocoders (Experimental)
We provide here different vocoder implementations which can be combined with our TTS models to enable "FASTER THAN REAL-TIME" end-to-end TTS stack.
Currently, there are implementations of the following models.
- Melgan
- MultiBand-Melgan
- GAN-TTS (Discriminator Only)
It is also very easy to adapt different vocoder models as we provide here a flexible and modular (but not too modular) framework.
## Training a model
You can see here an example (Soon)[Colab Notebook]() training MelGAN with LJSpeech dataset.
In order to train a new model, you need to collecto all your wav files under a common parent folder and give this path to `data_path` field in '''config.json'''
You need to define other relevant parameters in your ```config.json``` and then start traning with the following command from Mozilla TTS root path.
```CUDA_VISIBLE_DEVICES='1' python vocoder/train.py --config_path path/to/config.json```
Exampled config files can be found under `vocoder/configs/` folder.
You can continue a previous training by the following command.
```CUDA_VISIBLE_DEVICES='1' python vocoder/train.py --continue_path path/to/your/model/folder```
You can fine-tune a pre-trained model by the following command.
```CUDA_VISIBLE_DEVICES='1' python vocoder/train.py --restore_path path/to/your/model.pth.tar```
Restoring a model starts a new training in a different output folder. It only restores model weights with the given checkpoint file. However, continuing a training starts from the same conditions the previous training run left off.
You can also follow your training runs on Tensorboard as you do with our TTS models.
## Acknowledgement
Thanks to @kan-bayashi for his [repository](https://github.com/kan-bayashi/ParallelWaveGAN) being the start point of our work.

0
vocoder/__init__.py Normal file
View File

0
vocoder/compute_tts_features.py Normal file
View File

151
vocoder/configs/multiband-melgan_and_rwd_config.json Normal file
View File

@ -0,0 +1,151 @@
{
"run_name": "multiband-melgan-rwd",
"run_description": "multiband melgan with random window discriminator from https://arxiv.org/pdf/1909.11646.pdf",
// AUDIO PARAMETERS
"audio":{
// stft parameters
"num_freq": 513, // number of stft frequency levels. Size of the linear spectogram frame.
"win_length": 1024, // stft window length in ms.
"hop_length": 256, // stft window hop-lengh in ms.
"frame_length_ms": null, // stft window length in ms.If null, 'win_length' is used.
"frame_shift_ms": null, // stft window hop-lengh in ms. If null, 'hop_length' is used.
// Audio processing parameters
"sample_rate": 22050, // DATASET-RELATED: wav sample-rate. If different than the original data, it is resampled.
"preemphasis": 0.0, // pre-emphasis to reduce spec noise and make it more structured. If 0.0, no -pre-emphasis.
"ref_level_db": 20, // reference level db, theoretically 20db is the sound of air.
// Silence trimming
"do_trim_silence": true,// enable trimming of slience of audio as you load it. LJspeech (false), TWEB (false), Nancy (true)
"trim_db": 60, // threshold for timming silence. Set this according to your dataset.
// Griffin-Lim
"power": 1.5, // value to sharpen wav signals after GL algorithm.
"griffin_lim_iters": 60,// #griffin-lim iterations. 30-60 is a good range. Larger the value, slower the generation.
// MelSpectrogram parameters
"num_mels": 80, // size of the mel spec frame.
"mel_fmin": 0.0, // minimum freq level for mel-spec. ~50 for male and ~95 for female voices. Tune for dataset!!
"mel_fmax": 8000.0, // maximum freq level for mel-spec. Tune for dataset!!
// Normalization parameters
"signal_norm": true, // normalize spec values. Mean-Var normalization if 'stats_path' is defined otherwise range normalization defined by the other params.
"min_level_db": -100, // lower bound for normalization
"symmetric_norm": true, // move normalization to range [-1, 1]
"max_norm": 4.0, // scale normalization to range [-max_norm, max_norm] or [0, max_norm]
"clip_norm": true, // clip normalized values into the range.
"stats_path": null // DO NOT USE WITH MULTI_SPEAKER MODEL. scaler stats file computed by 'compute_statistics.py'. If it is defined, mean-std based notmalization is used and other normalization params are ignored
},
// DISTRIBUTED TRAINING
// "distributed":{
// "backend": "nccl",
// "url": "tcp:\/\/localhost:54321"
// },
// MODEL PARAMETERS
"use_pqmf": true,
// LOSS PARAMETERS
"use_stft_loss": true,
"use_subband_stft_loss": true,
"use_mse_gan_loss": true,
"use_hinge_gan_loss": false,
"use_feat_match_loss": false, // use only with melgan discriminators
// loss weights
"stft_loss_weight": 0.5,
"subband_stft_loss_weight": 0.5,
"mse_G_loss_weight": 2.5,
"hinge_G_loss_weight": 2.5,
"feat_match_loss_weight": 25,
// multiscale stft loss parameters
"stft_loss_params": {
"n_ffts": [1024, 2048, 512],
"hop_lengths": [120, 240, 50],
"win_lengths": [600, 1200, 240]
},
// subband multiscale stft loss parameters
"subband_stft_loss_params":{
"n_ffts": [384, 683, 171],
"hop_lengths": [30, 60, 10],
"win_lengths": [150, 300, 60]
},
"target_loss": "avg_G_loss", // loss value to pick the best model to save after each epoch
// DISCRIMINATOR
"discriminator_model": "random_window_discriminator",
"discriminator_model_params":{
"uncond_disc_donwsample_factors": [8, 4],
"cond_disc_downsample_factors": [[8, 4, 2, 2, 2], [8, 4, 2, 2], [8, 4, 2], [8, 4], [4, 2, 2]],
"cond_disc_out_channels": [[128, 128, 256, 256], [128, 256, 256], [128, 256], [256], [128, 256]],
"window_sizes": [512, 1024, 2048, 4096, 8192]
},
"steps_to_start_discriminator": 200000, // steps required to start GAN trainining.1
// GENERATOR
"generator_model": "multiband_melgan_generator",
"generator_model_params": {
"upsample_factors":[8, 4, 2],
"num_res_blocks": 4
},
// DATASET
"data_path": "/home/erogol/Data/LJSpeech-1.1/wavs/",
"seq_len": 16384,
"pad_short": 2000,
"conv_pad": 0,
"use_noise_augment": false,
"use_cache": true,
"reinit_layers": [], // give a list of layer names to restore from the given checkpoint. If not defined, it reloads all heuristically matching layers.
// TRAINING
"batch_size": 64, // Batch size for training. Lower values than 32 might cause hard to learn attention. It is overwritten by 'gradual_training'.
// VALIDATION
"run_eval": true,
"test_delay_epochs": 10, //Until attention is aligned, testing only wastes computation time.
"test_sentences_file": null, // set a file to load sentences to be used for testing. If it is null then we use default english sentences.
// OPTIMIZER
"noam_schedule": false, // use noam warmup and lr schedule.
"warmup_steps_gen": 4000, // Noam decay steps to increase the learning rate from 0 to "lr"
"warmup_steps_disc": 4000,
"epochs": 10000, // total number of epochs to train.
"wd": 0.0, // Weight decay weight.
"gen_clip_grad": -1, // Generator gradient clipping threshold. Apply gradient clipping if > 0
"disc_clip_grad": -1, // Discriminator gradient clipping threshold.
"lr_scheduler_gen": "MultiStepLR", // one of the schedulers from https://pytorch.org/docs/stable/optim.html#how-to-adjust-learning-rate
"lr_scheduler_gen_params": {
"gamma": 0.5,
"milestones": [100000, 200000, 300000, 400000, 500000, 600000]
},
"lr_scheduler_disc": "MultiStepLR", // one of the schedulers from https://pytorch.org/docs/stable/optim.html#how-to-adjust-learning-rate
"lr_scheduler_disc_params": {
"gamma": 0.5,
"milestones": [100000, 200000, 300000, 400000, 500000, 600000]
},
"lr_gen": 1e-4, // Initial learning rate. If Noam decay is active, maximum learning rate.
"lr_disc": 1e-4,
// TENSORBOARD and LOGGING
"print_step": 25, // Number of steps to log traning on console.
"print_eval": false, // If True, it prints loss values for each step in eval run.
"save_step": 25000, // Number of training steps expected to plot training stats on TB and save model checkpoints.
"checkpoint": true, // If true, it saves checkpoints per "save_step"
"tb_model_param_stats": false, // true, plots param stats per layer on tensorboard. Might be memory consuming, but good for debugging.
// DATA LOADING
"num_loader_workers": 4, // number of training data loader processes. Don't set it too big. 4-8 are good values.
"num_val_loader_workers": 4, // number of evaluation data loader processes.
"eval_split_size": 10,
// PATHS
"output_path": "/home/erogol/Models/LJSpeech/"
}

144
vocoder/configs/multiband_melgan_config.json Normal file
View File

@ -0,0 +1,144 @@
{
"run_name": "multiband-melgan",
"run_description": "multiband melgan mean-var scaling",
// AUDIO PARAMETERS
"audio":{
"fft_size": 1024, // number of stft frequency levels. Size of the linear spectogram frame.
"win_length": 1024, // stft window length in ms.
"hop_length": 256, // stft window hop-lengh in ms.
"frame_length_ms": null, // stft window length in ms.If null, 'win_length' is used.
"frame_shift_ms": null, // stft window hop-lengh in ms. If null, 'hop_length' is used.
// Audio processing parameters
"sample_rate": 22050, // DATASET-RELATED: wav sample-rate. If different than the original data, it is resampled.
"preemphasis": 0.0, // pre-emphasis to reduce spec noise and make it more structured. If 0.0, no -pre-emphasis.
"ref_level_db": 0, // reference level db, theoretically 20db is the sound of air.
// Silence trimming
"do_trim_silence": true,// enable trimming of slience of audio as you load it. LJspeech (false), TWEB (false), Nancy (true)
"trim_db": 60, // threshold for timming silence. Set this according to your dataset.
// MelSpectrogram parameters
"num_mels": 80, // size of the mel spec frame.
"mel_fmin": 50.0, // minimum freq level for mel-spec. ~50 for male and ~95 for female voices. Tune for dataset!!
"mel_fmax": 7600.0, // maximum freq level for mel-spec. Tune for dataset!!
"spec_gain": 1.0, // scaler value appplied after log transform of spectrogram.
// Normalization parameters
"signal_norm": true, // normalize spec values. Mean-Var normalization if 'stats_path' is defined otherwise range normalization defined by the other params.
"min_level_db": -100, // lower bound for normalization
"symmetric_norm": true, // move normalization to range [-1, 1]
"max_norm": 4.0, // scale normalization to range [-max_norm, max_norm] or [0, max_norm]
"clip_norm": true, // clip normalized values into the range.
"stats_path": "/home/erogol/Data/LJSpeech-1.1/scale_stats.npy" // DO NOT USE WITH MULTI_SPEAKER MODEL. scaler stats file computed by 'compute_statistics.py'. If it is defined, mean-std based notmalization is used and other normalization params are ignored
},
// DISTRIBUTED TRAINING
// "distributed":{
// "backend": "nccl",
// "url": "tcp:\/\/localhost:54321"
// },
// MODEL PARAMETERS
"use_pqmf": true,
// LOSS PARAMETERS
"use_stft_loss": true,
"use_subband_stft_loss": true,
"use_mse_gan_loss": true,
"use_hinge_gan_loss": false,
"use_feat_match_loss": false, // use only with melgan discriminators
// loss weights
"stft_loss_weight": 0.5,
"subband_stft_loss_weight": 0.5,
"mse_G_loss_weight": 2.5,
"hinge_G_loss_weight": 2.5,
"feat_match_loss_weight": 25,
// multiscale stft loss parameters
"stft_loss_params": {
"n_ffts": [1024, 2048, 512],
"hop_lengths": [120, 240, 50],
"win_lengths": [600, 1200, 240]
},
// subband multiscale stft loss parameters
"subband_stft_loss_params":{
"n_ffts": [384, 683, 171],
"hop_lengths": [30, 60, 10],
"win_lengths": [150, 300, 60]
},
"target_loss": "avg_G_loss", // loss value to pick the best model to save after each epoch
// DISCRIMINATOR
"discriminator_model": "melgan_multiscale_discriminator",
"discriminator_model_params":{
"base_channels": 16,
"max_channels":512,
"downsample_factors":[4, 4, 4]
},
"steps_to_start_discriminator": 200000, // steps required to start GAN trainining.1
// GENERATOR
"generator_model": "multiband_melgan_generator",
"generator_model_params": {
"upsample_factors":[8, 4, 2],
"num_res_blocks": 4
},
// DATASET
"data_path": "/home/erogol/Data/LJSpeech-1.1/wavs/",
"feature_path": null,
"seq_len": 16384,
"pad_short": 2000,
"conv_pad": 0,
"use_noise_augment": false,
"use_cache": true,
"reinit_layers": [], // give a list of layer names to restore from the given checkpoint. If not defined, it reloads all heuristically matching layers.
// TRAINING
"batch_size": 64, // Batch size for training. Lower values than 32 might cause hard to learn attention. It is overwritten by 'gradual_training'.
// VALIDATION
"run_eval": true,
"test_delay_epochs": 10, //Until attention is aligned, testing only wastes computation time.
"test_sentences_file": null, // set a file to load sentences to be used for testing. If it is null then we use default english sentences.
// OPTIMIZER
"epochs": 10000, // total number of epochs to train.
"wd": 0.0, // Weight decay weight.
"gen_clip_grad": -1, // Generator gradient clipping threshold. Apply gradient clipping if > 0
"disc_clip_grad": -1, // Discriminator gradient clipping threshold.
"lr_scheduler_gen": "MultiStepLR", // one of the schedulers from https://pytorch.org/docs/stable/optim.html#how-to-adjust-learning-rate
"lr_scheduler_gen_params": {
"gamma": 0.5,
"milestones": [100000, 200000, 300000, 400000, 500000, 600000]
},
"lr_scheduler_disc": "MultiStepLR", // one of the schedulers from https://pytorch.org/docs/stable/optim.html#how-to-adjust-learning-rate
"lr_scheduler_disc_params": {
"gamma": 0.5,
"milestones": [100000, 200000, 300000, 400000, 500000, 600000]
},
"lr_gen": 1e-4, // Initial learning rate. If Noam decay is active, maximum learning rate.
"lr_disc": 1e-4,
// TENSORBOARD and LOGGING
"print_step": 25, // Number of steps to log traning on console.
"print_eval": false, // If True, it prints loss values for each step in eval run.
"save_step": 25000, // Number of training steps expected to plot training stats on TB and save model checkpoints.
"checkpoint": true, // If true, it saves checkpoints per "save_step"
"tb_model_param_stats": false, // true, plots param stats per layer on tensorboard. Might be memory consuming, but good for debugging.
// DATA LOADING
"num_loader_workers": 4, // number of training data loader processes. Don't set it too big. 4-8 are good values.
"num_val_loader_workers": 4, // number of evaluation data loader processes.
"eval_split_size": 10,
// PATHS
"output_path": "/home/erogol/Models/LJSpeech/"
}

0
vocoder/datasets/__init__.py Normal file
View File

126
vocoder/datasets/gan_dataset.py Normal file
View File

@ -0,0 +1,126 @@
import os
import glob
import torch
import random
import numpy as np
from torch.utils.data import Dataset
from multiprocessing import Manager
class GANDataset(Dataset):
"""
GAN Dataset searchs for all the wav files under root path
and converts them to acoustic features on the fly and returns
random segments of (audio, feature) couples.
"""
def __init__(self,
ap,
items,
seq_len,
hop_len,
pad_short,
conv_pad=2,
is_training=True,
return_segments=True,
use_noise_augment=False,
use_cache=False,
verbose=False):
self.ap = ap
self.item_list = items
self.compute_feat = not isinstance(items[0], (tuple, list))
self.seq_len = seq_len
self.hop_len = hop_len
self.pad_short = pad_short
self.conv_pad = conv_pad
self.is_training = is_training
self.return_segments = return_segments
self.use_cache = use_cache
self.use_noise_augment = use_noise_augment
self.verbose = verbose
assert seq_len % hop_len == 0, " [!] seq_len has to be a multiple of hop_len."
self.feat_frame_len = seq_len // hop_len + (2 * conv_pad)
# map G and D instances
self.G_to_D_mappings = list(range(len(self.item_list)))
self.shuffle_mapping()
# cache acoustic features
if use_cache:
self.create_feature_cache()
def create_feature_cache(self):
self.manager = Manager()
self.cache = self.manager.list()
self.cache += [None for _ in range(len(self.item_list))]
@staticmethod
def find_wav_files(path):
return glob.glob(os.path.join(path, '**', '*.wav'), recursive=True)
def __len__(self):
return len(self.item_list)
def __getitem__(self, idx):
""" Return different items for Generator and Discriminator and
cache acoustic features """
if self.return_segments:
idx2 = self.G_to_D_mappings[idx]
item1 = self.load_item(idx)
item2 = self.load_item(idx2)
return item1, item2
item1 = self.load_item(idx)
return item1
def shuffle_mapping(self):
random.shuffle(self.G_to_D_mappings)
def load_item(self, idx):
""" load (audio, feat) couple """
if self.compute_feat:
# compute features from wav
wavpath = self.item_list[idx]
# print(wavpath)
if self.use_cache and self.cache[idx] is not None:
audio, mel = self.cache[idx]
else:
audio = self.ap.load_wav(wavpath)
mel = self.ap.melspectrogram(audio)
else:
# load precomputed features
wavpath, feat_path = self.item_list[idx]
if self.use_cache and self.cache[idx] is not None:
audio, mel = self.cache[idx]
else:
audio = self.ap.load_wav(wavpath)
mel = np.load(feat_path)
if len(audio) < self.seq_len + self.pad_short:
audio = np.pad(audio, (0, self.seq_len + self.pad_short - len(audio)), \
mode='constant', constant_values=0.0)
# correct the audio length wrt padding applied in stft
audio = np.pad(audio, (0, self.hop_len), mode="edge")
audio = audio[:mel.shape[-1] * self.hop_len]
assert mel.shape[-1] * self.hop_len == audio.shape[-1], f' [!] {mel.shape[-1] * self.hop_len} vs {audio.shape[-1]}'
audio = torch.from_numpy(audio).float().unsqueeze(0)
mel = torch.from_numpy(mel).float().squeeze(0)
if self.return_segments:
max_mel_start = mel.shape[1] - self.feat_frame_len
mel_start = random.randint(0, max_mel_start)
mel_end = mel_start + self.feat_frame_len
mel = mel[:, mel_start:mel_end]
audio_start = mel_start * self.hop_len
audio = audio[:, audio_start:audio_start +
self.seq_len]
if self.use_noise_augment and self.is_training and self.return_segments:
audio = audio + (1 / 32768) * torch.randn_like(audio)
return (mel, audio)

37
vocoder/datasets/preprocess.py Normal file
View File

@ -0,0 +1,37 @@
import glob
import os
from pathlib import Path
import numpy as np
def find_wav_files(data_path):
wav_paths = glob.glob(os.path.join(data_path, '**', '*.wav'), recursive=True)
return wav_paths
def find_feat_files(data_path):
feat_paths = glob.glob(os.path.join(data_path, '**', '*.npy'), recursive=True)
return feat_paths
def load_wav_data(data_path, eval_split_size):
wav_paths = find_wav_files(data_path)
np.random.seed(0)
np.random.shuffle(wav_paths)
return wav_paths[:eval_split_size], wav_paths[eval_split_size:]
def load_wav_feat_data(data_path, feat_path, eval_split_size):
wav_paths = sorted(find_wav_files(data_path))
feat_paths = sorted(find_feat_files(feat_path))
assert len(wav_paths) == len(feat_paths)
for wav, feat in zip(wav_paths, feat_paths):
wav_name = Path(wav).stem
feat_name = Path(feat).stem
assert wav_name == feat_name
items = list(zip(wav_paths, feat_paths))
np.random.seed(0)
np.random.shuffle(items)
return items[:eval_split_size], items[eval_split_size:]

0
vocoder/layers/__init__.py Normal file
View File

309
vocoder/layers/losses.py Normal file
View File

@ -0,0 +1,309 @@
import torch
from torch import nn
from torch.nn import functional as F
class TorchSTFT():
def __init__(self, n_fft, hop_length, win_length, window='hann_window'):
""" Torch based STFT operation """
self.n_fft = n_fft
self.hop_length = hop_length
self.win_length = win_length
self.window = getattr(torch, window)(win_length)
def __call__(self, x):
# B x D x T x 2
o = torch.stft(x,
self.n_fft,
self.hop_length,
self.win_length,
self.window,
center=True,
pad_mode="reflect", # compatible with audio.py
normalized=False,
onesided=True)
M = o[:, :, :, 0]
P = o[:, :, :, 1]
return torch.sqrt(torch.clamp(M ** 2 + P ** 2, min=1e-8))
#################################
# GENERATOR LOSSES
#################################
class STFTLoss(nn.Module):
""" Single scale STFT Loss """
def __init__(self, n_fft, hop_length, win_length):
super(STFTLoss, self).__init__()
self.n_fft = n_fft
self.hop_length = hop_length
self.win_length = win_length
self.stft = TorchSTFT(n_fft, hop_length, win_length)
def forward(self, y_hat, y):
y_hat_M = self.stft(y_hat)
y_M = self.stft(y)
# magnitude loss
loss_mag = F.l1_loss(torch.log(y_M), torch.log(y_hat_M))
# spectral convergence loss
loss_sc = torch.norm(y_M - y_hat_M, p="fro") / torch.norm(y_M, p="fro")
return loss_mag, loss_sc
class MultiScaleSTFTLoss(torch.nn.Module):
""" Multi scale STFT loss """
def __init__(self,
n_ffts=(1024, 2048, 512),
hop_lengths=(120, 240, 50),
win_lengths=(600, 1200, 240)):
super(MultiScaleSTFTLoss, self).__init__()
self.loss_funcs = torch.nn.ModuleList()
for n_fft, hop_length, win_length in zip(n_ffts, hop_lengths, win_lengths):
self.loss_funcs.append(STFTLoss(n_fft, hop_length, win_length))
def forward(self, y_hat, y):
N = len(self.loss_funcs)
loss_sc = 0
loss_mag = 0
for f in self.loss_funcs:
lm, lsc = f(y_hat, y)
loss_mag += lm
loss_sc += lsc
loss_sc /= N
loss_mag /= N
return loss_mag, loss_sc
class MultiScaleSubbandSTFTLoss(MultiScaleSTFTLoss):
""" Multiscale STFT loss for multi band model outputs """
# pylint: disable=no-self-use
def forward(self, y_hat, y):
y_hat = y_hat.view(-1, 1, y_hat.shape[2])
y = y.view(-1, 1, y.shape[2])
return super().forward(y_hat.squeeze(1), y.squeeze(1))
class MSEGLoss(nn.Module):
""" Mean Squared Generator Loss """
# pylint: disable=no-self-use
def forward(self, score_real):
loss_fake = F.mse_loss(score_real, score_real.new_ones(score_real.shape))
return loss_fake
class HingeGLoss(nn.Module):
""" Hinge Discriminator Loss """
# pylint: disable=no-self-use
def forward(self, score_real):
# TODO: this might be wrong
loss_fake = torch.mean(F.relu(1. - score_real))
return loss_fake
##################################
# DISCRIMINATOR LOSSES
##################################
class MSEDLoss(nn.Module):
""" Mean Squared Discriminator Loss """
def __init__(self,):
super(MSEDLoss, self).__init__()
self.loss_func = nn.MSELoss()
# pylint: disable=no-self-use
def forward(self, score_fake, score_real):
loss_real = self.loss_func(score_real, score_real.new_ones(score_real.shape))
loss_fake = self.loss_func(score_fake, score_fake.new_zeros(score_fake.shape))
loss_d = loss_real + loss_fake
return loss_d, loss_real, loss_fake
class HingeDLoss(nn.Module):
""" Hinge Discriminator Loss """
# pylint: disable=no-self-use
def forward(self, score_fake, score_real):
loss_real = torch.mean(F.relu(1. - score_real))
loss_fake = torch.mean(F.relu(1. + score_fake))
loss_d = loss_real + loss_fake
return loss_d, loss_real, loss_fake
class MelganFeatureLoss(nn.Module):
def __init__(self,):
super(MelganFeatureLoss, self).__init__()
self.loss_func = nn.L1Loss()
# pylint: disable=no-self-use
def forward(self, fake_feats, real_feats):
loss_feats = 0
for fake_feat, real_feat in zip(fake_feats, real_feats):
loss_feats += self.loss_func(fake_feat, real_feat)
loss_feats /= len(fake_feats) + len(real_feats)
return loss_feats
#####################################
# LOSS WRAPPERS
#####################################
def _apply_G_adv_loss(scores_fake, loss_func):
""" Compute G adversarial loss function
and normalize values """
adv_loss = 0
if isinstance(scores_fake, list):
for score_fake in scores_fake:
fake_loss = loss_func(score_fake)
adv_loss += fake_loss
adv_loss /= len(scores_fake)
else:
fake_loss = loss_func(scores_fake)
adv_loss = fake_loss
return adv_loss
def _apply_D_loss(scores_fake, scores_real, loss_func):
""" Compute D loss func and normalize loss values """
loss = 0
real_loss = 0
fake_loss = 0
if isinstance(scores_fake, list):
# multi-scale loss
for score_fake, score_real in zip(scores_fake, scores_real):
total_loss, real_loss, fake_loss = loss_func(score_fake=score_fake, score_real=score_real)
loss += total_loss
real_loss += real_loss
fake_loss += fake_loss
# normalize loss values with number of scales
loss /= len(scores_fake)
real_loss /= len(scores_real)
fake_loss /= len(scores_fake)
else:
# single scale loss
total_loss, real_loss, fake_loss = loss_func(scores_fake, scores_real)
loss = total_loss
return loss, real_loss, fake_loss
##################################
# MODEL LOSSES
##################################
class GeneratorLoss(nn.Module):
def __init__(self, C):
""" Compute Generator Loss values depending on training
configuration """
super(GeneratorLoss, self).__init__()
assert not(C.use_mse_gan_loss and C.use_hinge_gan_loss),\
" [!] Cannot use HingeGANLoss and MSEGANLoss together."
self.use_stft_loss = C.use_stft_loss
self.use_subband_stft_loss = C.use_subband_stft_loss
self.use_mse_gan_loss = C.use_mse_gan_loss
self.use_hinge_gan_loss = C.use_hinge_gan_loss
self.use_feat_match_loss = C.use_feat_match_loss
self.stft_loss_weight = C.stft_loss_weight
self.subband_stft_loss_weight = C.subband_stft_loss_weight
self.mse_gan_loss_weight = C.mse_G_loss_weight
self.hinge_gan_loss_weight = C.hinge_G_loss_weight
self.feat_match_loss_weight = C.feat_match_loss_weight
if C.use_stft_loss:
self.stft_loss = MultiScaleSTFTLoss(**C.stft_loss_params)
if C.use_subband_stft_loss:
self.subband_stft_loss = MultiScaleSubbandSTFTLoss(**C.subband_stft_loss_params)
if C.use_mse_gan_loss:
self.mse_loss = MSEGLoss()
if C.use_hinge_gan_loss:
self.hinge_loss = HingeGLoss()
if C.use_feat_match_loss:
self.feat_match_loss = MelganFeatureLoss()
def forward(self, y_hat=None, y=None, scores_fake=None, feats_fake=None, feats_real=None, y_hat_sub=None, y_sub=None):
gen_loss = 0
adv_loss = 0
return_dict = {}
# STFT Loss
if self.use_stft_loss:
stft_loss_mg, stft_loss_sc = self.stft_loss(y_hat.squeeze(1), y.squeeze(1))
return_dict['G_stft_loss_mg'] = stft_loss_mg
return_dict['G_stft_loss_sc'] = stft_loss_sc
gen_loss += self.stft_loss_weight * (stft_loss_mg + stft_loss_sc)
# subband STFT Loss
if self.use_subband_stft_loss:
subband_stft_loss_mg, subband_stft_loss_sc = self.subband_stft_loss(y_hat_sub, y_sub)
return_dict['G_subband_stft_loss_mg'] = subband_stft_loss_mg
return_dict['G_subband_stft_loss_sc'] = subband_stft_loss_sc
gen_loss += self.subband_stft_loss_weight * (subband_stft_loss_mg + subband_stft_loss_sc)
# multiscale MSE adversarial loss
if self.use_mse_gan_loss and scores_fake is not None:
mse_fake_loss = _apply_G_adv_loss(scores_fake, self.mse_loss)
return_dict['G_mse_fake_loss'] = mse_fake_loss
adv_loss += self.mse_gan_loss_weight * mse_fake_loss
# multiscale Hinge adversarial loss
if self.use_hinge_gan_loss and not scores_fake is not None:
hinge_fake_loss = _apply_G_adv_loss(scores_fake, self.hinge_loss)
return_dict['G_hinge_fake_loss'] = hinge_fake_loss
adv_loss += self.hinge_gan_loss_weight * hinge_fake_loss
# Feature Matching Loss
if self.use_feat_match_loss and not feats_fake:
feat_match_loss = self.feat_match_loss(feats_fake, feats_real)
return_dict['G_feat_match_loss'] = feat_match_loss
adv_loss += self.feat_match_loss_weight * feat_match_loss
return_dict['G_loss'] = gen_loss + adv_loss
return_dict['G_gen_loss'] = gen_loss
return_dict['G_adv_loss'] = adv_loss
return return_dict
class DiscriminatorLoss(nn.Module):
""" Compute Discriminator Loss values depending on training
configuration """
def __init__(self, C):
super(DiscriminatorLoss, self).__init__()
assert not(C.use_mse_gan_loss and C.use_hinge_gan_loss),\
" [!] Cannot use HingeGANLoss and MSEGANLoss together."
self.use_mse_gan_loss = C.use_mse_gan_loss
self.use_hinge_gan_loss = C.use_hinge_gan_loss
if C.use_mse_gan_loss:
self.mse_loss = MSEDLoss()
if C.use_hinge_gan_loss:
self.hinge_loss = HingeDLoss()
def forward(self, scores_fake, scores_real):
loss = 0
return_dict = {}
if self.use_mse_gan_loss:
mse_D_loss, mse_D_real_loss, mse_D_fake_loss = _apply_D_loss(
scores_fake=scores_fake,
scores_real=scores_real,
loss_func=self.mse_loss)
return_dict['D_mse_gan_loss'] = mse_D_loss
return_dict['D_mse_gan_real_loss'] = mse_D_real_loss
return_dict['D_mse_gan_fake_loss'] = mse_D_fake_loss
loss += mse_D_loss
if self.use_hinge_gan_loss:
hinge_D_loss, hinge_D_real_loss, hinge_D_fake_loss = _apply_D_loss(
scores_fake=scores_fake,
scores_real=scores_real,
loss_func=self.hinge_loss)
return_dict['D_hinge_gan_loss'] = hinge_D_loss
return_dict['D_hinge_gan_real_loss'] = hinge_D_real_loss
return_dict['D_hinge_gan_fake_loss'] = hinge_D_fake_loss
loss += hinge_D_loss
return_dict['D_loss'] = loss
return return_dict

45
vocoder/layers/melgan.py Normal file
View File

@ -0,0 +1,45 @@
from torch import nn
from torch.nn.utils import weight_norm
class ResidualStack(nn.Module):
def __init__(self, channels, num_res_blocks, kernel_size):
super(ResidualStack, self).__init__()
assert (kernel_size - 1) % 2 == 0, " [!] kernel_size has to be odd."
base_padding = (kernel_size - 1) // 2
self.blocks = nn.ModuleList()
for idx in range(num_res_blocks):
layer_kernel_size = kernel_size
layer_dilation = layer_kernel_size**idx
layer_padding = base_padding * layer_dilation
self.blocks += [nn.Sequential(
nn.LeakyReLU(0.2),
nn.ReflectionPad1d(layer_padding),
weight_norm(
nn.Conv1d(channels,
channels,
kernel_size=kernel_size,
dilation=layer_dilation,
bias=True)),
nn.LeakyReLU(0.2),
weight_norm(
nn.Conv1d(channels, channels, kernel_size=1, bias=True)),
)]
self.shortcuts = nn.ModuleList([
weight_norm(nn.Conv1d(channels, channels, kernel_size=1,
bias=True)) for i in range(num_res_blocks)
])
def forward(self, x):
for block, shortcut in zip(self.blocks, self.shortcuts):
x = shortcut(x) + block(x)
return x
def remove_weight_norm(self):
for block, shortcut in zip(self.blocks, self.shortcuts):
nn.utils.remove_weight_norm(block[2])
nn.utils.remove_weight_norm(block[4])
nn.utils.remove_weight_norm(shortcut)

56
vocoder/layers/pqmf.py Normal file
View File

@ -0,0 +1,56 @@
import numpy as np
import torch
import torch.nn.functional as F
from scipy import signal as sig
# adapted from
# https://github.com/kan-bayashi/ParallelWaveGAN/tree/master/parallel_wavegan
class PQMF(torch.nn.Module):
def __init__(self, N=4, taps=62, cutoff=0.15, beta=9.0):
super(PQMF, self).__init__()
self.N = N
self.taps = taps
self.cutoff = cutoff
self.beta = beta
QMF = sig.firwin(taps + 1, cutoff, window=('kaiser', beta))
H = np.zeros((N, len(QMF)))
G = np.zeros((N, len(QMF)))
for k in range(N):
constant_factor = (2 * k + 1) * (np.pi /
(2 * N)) * (np.arange(taps + 1) -
((taps - 1) / 2))
phase = (-1)**k * np.pi / 4
H[k] = 2 * QMF * np.cos(constant_factor + phase)
G[k] = 2 * QMF * np.cos(constant_factor - phase)
H = torch.from_numpy(H[:, None, :]).float()
G = torch.from_numpy(G[None, :, :]).float()
self.register_buffer("H", H)
self.register_buffer("G", G)
updown_filter = torch.zeros((N, N, N)).float()
for k in range(N):
updown_filter[k, k, 0] = 1.0
self.register_buffer("updown_filter", updown_filter)
self.N = N
self.pad_fn = torch.nn.ConstantPad1d(taps // 2, 0.0)
def forward(self, x):
return self.analysis(x)
def analysis(self, x):
return F.conv1d(x, self.H, padding=self.taps // 2, stride=self.N)
def synthesis(self, x):
x = F.conv_transpose1d(x,
self.updown_filter * self.N,
stride=self.N)
x = F.conv1d(x, self.G, padding=self.taps // 2)
return x

640
vocoder/layers/qmf.dat Normal file
View File

@ -0,0 +1,640 @@
0.0000000e+000
-5.5252865e-004
-5.6176926e-004
-4.9475181e-004
-4.8752280e-004
-4.8937912e-004
-5.0407143e-004
-5.2265643e-004
-5.4665656e-004
-5.6778026e-004
-5.8709305e-004
-6.1327474e-004
-6.3124935e-004
-6.5403334e-004
-6.7776908e-004
-6.9416146e-004
-7.1577365e-004
-7.2550431e-004
-7.4409419e-004
-7.4905981e-004
-7.6813719e-004
-7.7248486e-004
-7.8343323e-004
-7.7798695e-004
-7.8036647e-004
-7.8014496e-004
-7.7579773e-004
-7.6307936e-004
-7.5300014e-004
-7.3193572e-004
-7.2153920e-004
-6.9179375e-004
-6.6504151e-004
-6.3415949e-004
-5.9461189e-004
-5.5645764e-004
-5.1455722e-004
-4.6063255e-004
-4.0951215e-004
-3.5011759e-004
-2.8969812e-004
-2.0983373e-004
-1.4463809e-004
-6.1733441e-005
1.3494974e-005
1.0943831e-004
2.0430171e-004
2.9495311e-004
4.0265402e-004
5.1073885e-004
6.2393761e-004
7.4580259e-004
8.6084433e-004
9.8859883e-004
1.1250155e-003
1.2577885e-003
1.3902495e-003
1.5443220e-003
1.6868083e-003
1.8348265e-003
1.9841141e-003
2.1461584e-003
2.3017255e-003
2.4625617e-003
2.6201759e-003
2.7870464e-003
2.9469448e-003
3.1125421e-003
3.2739613e-003
3.4418874e-003
3.6008268e-003
3.7603923e-003
3.9207432e-003
4.0819753e-003
4.2264269e-003
4.3730720e-003
4.5209853e-003
4.6606461e-003
4.7932561e-003
4.9137604e-003
5.0393023e-003
5.1407354e-003
5.2461166e-003
5.3471681e-003
5.4196776e-003
5.4876040e-003
5.5475715e-003
5.5938023e-003
5.6220643e-003
5.6455197e-003
5.6389200e-003
5.6266114e-003
5.5917129e-003
5.5404364e-003
5.4753783e-003
5.3838976e-003
5.2715759e-003
5.1382275e-003
4.9839688e-003
4.8109469e-003
4.6039530e-003
4.3801862e-003
4.1251642e-003
3.8456408e-003
3.5401247e-003
3.2091886e-003
2.8446758e-003
2.4508540e-003
2.0274176e-003
1.5784683e-003
1.0902329e-003
5.8322642e-004
2.7604519e-005
-5.4642809e-004
-1.1568136e-003
-1.8039473e-003
-2.4826724e-003
-3.1933778e-003
-3.9401124e-003
-4.7222596e-003
-5.5337211e-003
-6.3792293e-003
-7.2615817e-003
-8.1798233e-003
-9.1325330e-003
-1.0115022e-002
-1.1131555e-002
-1.2185000e-002
-1.3271822e-002
-1.4390467e-002
-1.5540555e-002
-1.6732471e-002
-1.7943338e-002
-1.9187243e-002
-2.0453179e-002
-2.1746755e-002
-2.3068017e-002
-2.4416099e-002
-2.5787585e-002
-2.7185943e-002
-2.8607217e-002
-3.0050266e-002
-3.1501761e-002
-3.2975408e-002
-3.4462095e-002
-3.5969756e-002
-3.7481285e-002
-3.9005368e-002
-4.0534917e-002
-4.2064909e-002
-4.3609754e-002
-4.5148841e-002
-4.6684303e-002
-4.8216572e-002
-4.9738576e-002
-5.1255616e-002
-5.2763075e-002
-5.4245277e-002
-5.5717365e-002
-5.7161645e-002
-5.8591568e-002
-5.9983748e-002
-6.1345517e-002
-6.2685781e-002
-6.3971590e-002
-6.5224711e-002
-6.6436751e-002
-6.7607599e-002
-6.8704383e-002
-6.9763024e-002
-7.0762871e-002
-7.1700267e-002
-7.2568258e-002
-7.3362026e-002
-7.4100364e-002
-7.4745256e-002
-7.5313734e-002
-7.5800836e-002
-7.6199248e-002
-7.6499217e-002
-7.6709349e-002
-7.6817398e-002
-7.6823001e-002
-7.6720492e-002
-7.6505072e-002
-7.6174832e-002
-7.5730576e-002
-7.5157626e-002
-7.4466439e-002
-7.3640601e-002
-7.2677464e-002
-7.1582636e-002
-7.0353307e-002
-6.8966401e-002
-6.7452502e-002
-6.5769067e-002
-6.3944481e-002
-6.1960278e-002
-5.9816657e-002
-5.7515269e-002
-5.5046003e-002
-5.2409382e-002
-4.9597868e-002
-4.6630331e-002
-4.3476878e-002
-4.0145828e-002
-3.6641812e-002
-3.2958393e-002
-2.9082401e-002
-2.5030756e-002
-2.0799707e-002
-1.6370126e-002
-1.1762383e-002
-6.9636862e-003
-1.9765601e-003
3.2086897e-003
8.5711749e-003
1.4128883e-002
1.9883413e-002
2.5822729e-002
3.1953127e-002
3.8277657e-002
4.4780682e-002
5.1480418e-002
5.8370533e-002
6.5440985e-002
7.2694330e-002
8.0137293e-002
8.7754754e-002
9.5553335e-002
1.0353295e-001
1.1168269e-001
1.2000780e-001
1.2850029e-001
1.3715518e-001
1.4597665e-001
1.5496071e-001
1.6409589e-001
1.7338082e-001
1.8281725e-001
1.9239667e-001
2.0212502e-001
2.1197359e-001
2.2196527e-001
2.3206909e-001
2.4230169e-001
2.5264803e-001
2.6310533e-001
2.7366340e-001
2.8432142e-001
2.9507167e-001
3.0590986e-001
3.1682789e-001
3.2781137e-001
3.3887227e-001
3.4999141e-001
3.6115899e-001
3.7237955e-001
3.8363500e-001
3.9492118e-001
4.0623177e-001
4.1756969e-001
4.2891199e-001
4.4025538e-001
4.5159965e-001
4.6293081e-001
4.7424532e-001
4.8552531e-001
4.9677083e-001
5.0798175e-001
5.1912350e-001
5.3022409e-001
5.4125534e-001
5.5220513e-001
5.6307891e-001
5.7385241e-001
5.8454032e-001
5.9511231e-001
6.0557835e-001
6.1591099e-001
6.2612427e-001
6.3619801e-001
6.4612697e-001
6.5590163e-001
6.6551399e-001
6.7496632e-001
6.8423533e-001
6.9332824e-001
7.0223887e-001
7.1094104e-001
7.1944626e-001
7.2774489e-001
7.3582118e-001
7.4368279e-001
7.5131375e-001
7.5870808e-001
7.6586749e-001
7.7277809e-001
7.7942875e-001
7.8583531e-001
7.9197358e-001
7.9784664e-001
8.0344858e-001
8.0876950e-001
8.1381913e-001
8.1857760e-001
8.2304199e-001
8.2722753e-001
8.3110385e-001
8.3469374e-001
8.3797173e-001
8.4095414e-001
8.4362383e-001
8.4598185e-001
8.4803158e-001
8.4978052e-001
8.5119715e-001
8.5230470e-001
8.5310209e-001
8.5357206e-001
8.5373856e-001
8.5357206e-001
8.5310209e-001
8.5230470e-001
8.5119715e-001
8.4978052e-001
8.4803158e-001
8.4598185e-001
8.4362383e-001
8.4095414e-001
8.3797173e-001
8.3469374e-001
8.3110385e-001
8.2722753e-001
8.2304199e-001
8.1857760e-001
8.1381913e-001
8.0876950e-001
8.0344858e-001
7.9784664e-001
7.9197358e-001
7.8583531e-001
7.7942875e-001
7.7277809e-001
7.6586749e-001
7.5870808e-001
7.5131375e-001
7.4368279e-001
7.3582118e-001
7.2774489e-001
7.1944626e-001
7.1094104e-001
7.0223887e-001
6.9332824e-001
6.8423533e-001
6.7496632e-001
6.6551399e-001
6.5590163e-001
6.4612697e-001
6.3619801e-001
6.2612427e-001
6.1591099e-001
6.0557835e-001
5.9511231e-001
5.8454032e-001
5.7385241e-001
5.6307891e-001
5.5220513e-001
5.4125534e-001
5.3022409e-001
5.1912350e-001
5.0798175e-001
4.9677083e-001
4.8552531e-001
4.7424532e-001
4.6293081e-001
4.5159965e-001
4.4025538e-001
4.2891199e-001
4.1756969e-001
4.0623177e-001
3.9492118e-001
3.8363500e-001
3.7237955e-001
3.6115899e-001
3.4999141e-001
3.3887227e-001
3.2781137e-001
3.1682789e-001
3.0590986e-001
2.9507167e-001
2.8432142e-001
2.7366340e-001
2.6310533e-001
2.5264803e-001
2.4230169e-001
2.3206909e-001
2.2196527e-001
2.1197359e-001
2.0212502e-001
1.9239667e-001
1.8281725e-001
1.7338082e-001
1.6409589e-001
1.5496071e-001
1.4597665e-001
1.3715518e-001
1.2850029e-001
1.2000780e-001
1.1168269e-001
1.0353295e-001
9.5553335e-002
8.7754754e-002
8.0137293e-002
7.2694330e-002
6.5440985e-002
5.8370533e-002
5.1480418e-002
4.4780682e-002
3.8277657e-002
3.1953127e-002
2.5822729e-002
1.9883413e-002
1.4128883e-002
8.5711749e-003
3.2086897e-003
-1.9765601e-003
-6.9636862e-003
-1.1762383e-002
-1.6370126e-002
-2.0799707e-002
-2.5030756e-002
-2.9082401e-002
-3.2958393e-002
-3.6641812e-002
-4.0145828e-002
-4.3476878e-002
-4.6630331e-002
-4.9597868e-002
-5.2409382e-002
-5.5046003e-002
-5.7515269e-002
-5.9816657e-002
-6.1960278e-002
-6.3944481e-002
-6.5769067e-002
-6.7452502e-002
-6.8966401e-002
-7.0353307e-002
-7.1582636e-002
-7.2677464e-002
-7.3640601e-002
-7.4466439e-002
-7.5157626e-002
-7.5730576e-002
-7.6174832e-002
-7.6505072e-002
-7.6720492e-002
-7.6823001e-002
-7.6817398e-002
-7.6709349e-002
-7.6499217e-002
-7.6199248e-002
-7.5800836e-002
-7.5313734e-002
-7.4745256e-002
-7.4100364e-002
-7.3362026e-002
-7.2568258e-002
-7.1700267e-002
-7.0762871e-002
-6.9763024e-002
-6.8704383e-002
-6.7607599e-002
-6.6436751e-002
-6.5224711e-002
-6.3971590e-002
-6.2685781e-002
-6.1345517e-002
-5.9983748e-002
-5.8591568e-002
-5.7161645e-002
-5.5717365e-002
-5.4245277e-002
-5.2763075e-002
-5.1255616e-002
-4.9738576e-002
-4.8216572e-002
-4.6684303e-002
-4.5148841e-002
-4.3609754e-002
-4.2064909e-002
-4.0534917e-002
-3.9005368e-002
-3.7481285e-002
-3.5969756e-002
-3.4462095e-002
-3.2975408e-002
-3.1501761e-002
-3.0050266e-002
-2.8607217e-002
-2.7185943e-002
-2.5787585e-002
-2.4416099e-002
-2.3068017e-002
-2.1746755e-002
-2.0453179e-002
-1.9187243e-002
-1.7943338e-002
-1.6732471e-002
-1.5540555e-002
-1.4390467e-002
-1.3271822e-002
-1.2185000e-002
-1.1131555e-002
-1.0115022e-002
-9.1325330e-003
-8.1798233e-003
-7.2615817e-003
-6.3792293e-003
-5.5337211e-003
-4.7222596e-003
-3.9401124e-003
-3.1933778e-003
-2.4826724e-003
-1.8039473e-003
-1.1568136e-003
-5.4642809e-004
2.7604519e-005
5.8322642e-004
1.0902329e-003
1.5784683e-003
2.0274176e-003
2.4508540e-003
2.8446758e-003
3.2091886e-003
3.5401247e-003
3.8456408e-003
4.1251642e-003
4.3801862e-003
4.6039530e-003
4.8109469e-003
4.9839688e-003
5.1382275e-003
5.2715759e-003
5.3838976e-003
5.4753783e-003
5.5404364e-003
5.5917129e-003
5.6266114e-003
5.6389200e-003
5.6455197e-003
5.6220643e-003
5.5938023e-003
5.5475715e-003
5.4876040e-003
5.4196776e-003
5.3471681e-003
5.2461166e-003
5.1407354e-003
5.0393023e-003
4.9137604e-003
4.7932561e-003
4.6606461e-003
4.5209853e-003
4.3730720e-003
4.2264269e-003
4.0819753e-003
3.9207432e-003
3.7603923e-003
3.6008268e-003
3.4418874e-003
3.2739613e-003
3.1125421e-003
2.9469448e-003
2.7870464e-003
2.6201759e-003
2.4625617e-003
2.3017255e-003
2.1461584e-003
1.9841141e-003
1.8348265e-003
1.6868083e-003
1.5443220e-003
1.3902495e-003
1.2577885e-003
1.1250155e-003
9.8859883e-004
8.6084433e-004
7.4580259e-004
6.2393761e-004
5.1073885e-004
4.0265402e-004
2.9495311e-004
2.0430171e-004
1.0943831e-004
1.3494974e-005
-6.1733441e-005
-1.4463809e-004
-2.0983373e-004
-2.8969812e-004
-3.5011759e-004
-4.0951215e-004
-4.6063255e-004
-5.1455722e-004
-5.5645764e-004
-5.9461189e-004
-6.3415949e-004
-6.6504151e-004
-6.9179375e-004
-7.2153920e-004
-7.3193572e-004
-7.5300014e-004
-7.6307936e-004
-7.7579773e-004
-7.8014496e-004
-7.8036647e-004
-7.7798695e-004
-7.8343323e-004
-7.7248486e-004
-7.6813719e-004
-7.4905981e-004
-7.4409419e-004
-7.2550431e-004
-7.1577365e-004
-6.9416146e-004
-6.7776908e-004
-6.5403334e-004
-6.3124935e-004
-6.1327474e-004
-5.8709305e-004
-5.6778026e-004
-5.4665656e-004
-5.2265643e-004
-5.0407143e-004
-4.8937912e-004
-4.8752280e-004
-4.9475181e-004
-5.6176926e-004
-5.5252865e-004

0
vocoder/models/__init__.py Normal file
View File

78
vocoder/models/melgan_discriminator.py Normal file
View File

@ -0,0 +1,78 @@
import numpy as np
from torch import nn
from torch.nn.utils import weight_norm
class MelganDiscriminator(nn.Module):
def __init__(self,
in_channels=1,
out_channels=1,
kernel_sizes=(5, 3),
base_channels=16,
max_channels=1024,
downsample_factors=(4, 4, 4, 4)):
super(MelganDiscriminator, self).__init__()
self.layers = nn.ModuleList()
layer_kernel_size = np.prod(kernel_sizes)
layer_padding = (layer_kernel_size - 1) // 2
# initial layer
self.layers += [
nn.Sequential(
nn.ReflectionPad1d(layer_padding),
weight_norm(
nn.Conv1d(in_channels,
base_channels,
layer_kernel_size,
stride=1)), nn.LeakyReLU(0.2, inplace=True))
]
# downsampling layers
layer_in_channels = base_channels
for downsample_factor in downsample_factors:
layer_out_channels = min(layer_in_channels * downsample_factor,
max_channels)
layer_kernel_size = downsample_factor * 10 + 1
layer_padding = (layer_kernel_size - 1) // 2
layer_groups = layer_in_channels // 4
self.layers += [
nn.Sequential(
weight_norm(
nn.Conv1d(layer_in_channels,
layer_out_channels,
kernel_size=layer_kernel_size,
stride=downsample_factor,
padding=layer_padding,
groups=layer_groups)),
nn.LeakyReLU(0.2, inplace=True))
]
layer_in_channels = layer_out_channels
# last 2 layers
layer_padding1 = (kernel_sizes[0] - 1) // 2
layer_padding2 = (kernel_sizes[1] - 1) // 2
self.layers += [
nn.Sequential(
weight_norm(
nn.Conv1d(layer_out_channels,
layer_out_channels,
kernel_size=kernel_sizes[0],
stride=1,
padding=layer_padding1)),
nn.LeakyReLU(0.2, inplace=True),
),
weight_norm(
nn.Conv1d(layer_out_channels,
out_channels,
kernel_size=kernel_sizes[1],
stride=1,
padding=layer_padding2)),
]
def forward(self, x):
feats = []
for layer in self.layers:
x = layer(x)
feats.append(x)
return x, feats

98
vocoder/models/melgan_generator.py Normal file
View File

@ -0,0 +1,98 @@
import torch
from torch import nn
from torch.nn.utils import weight_norm
from TTS.vocoder.layers.melgan import ResidualStack
class MelganGenerator(nn.Module):
def __init__(self,
in_channels=80,
out_channels=1,
proj_kernel=7,
base_channels=512,
upsample_factors=(8, 8, 2, 2),
res_kernel=3,
num_res_blocks=3):
super(MelganGenerator, self).__init__()
# assert model parameters
assert (proj_kernel -
1) % 2 == 0, " [!] proj_kernel should be an odd number."
# setup additional model parameters
base_padding = (proj_kernel - 1) // 2
act_slope = 0.2
self.inference_padding = 2
# initial layer
layers = []
layers += [
nn.ReflectionPad1d(base_padding),
weight_norm(
nn.Conv1d(in_channels,
base_channels,
kernel_size=proj_kernel,
stride=1,
bias=True))
]
# upsampling layers and residual stacks
for idx, upsample_factor in enumerate(upsample_factors):
layer_in_channels = base_channels // (2**idx)
layer_out_channels = base_channels // (2**(idx + 1))
layer_filter_size = upsample_factor * 2
layer_stride = upsample_factor
layer_output_padding = upsample_factor % 2
layer_padding = upsample_factor // 2 + layer_output_padding
layers += [
nn.LeakyReLU(act_slope),
weight_norm(
nn.ConvTranspose1d(layer_in_channels,
layer_out_channels,
layer_filter_size,
stride=layer_stride,
padding=layer_padding,
output_padding=layer_output_padding,
bias=True)),
ResidualStack(
channels=layer_out_channels,
num_res_blocks=num_res_blocks,
kernel_size=res_kernel
)
]
layers += [nn.LeakyReLU(act_slope)]
# final layer
layers += [
nn.ReflectionPad1d(base_padding),
weight_norm(
nn.Conv1d(layer_out_channels,
out_channels,
proj_kernel,
stride=1,
bias=True)),
nn.Tanh()
]
self.layers = nn.Sequential(*layers)
def forward(self, c):
return self.layers(c)
def inference(self, c):
c = c.to(self.layers[1].weight.device)
c = torch.nn.functional.pad(
c,
(self.inference_padding, self.inference_padding),
'replicate')
return self.layers(c)
def remove_weight_norm(self):
for _, layer in enumerate(self.layers):
if len(layer.state_dict()) != 0:
try:
nn.utils.remove_weight_norm(layer)
except ValueError:
layer.remove_weight_norm()

41
vocoder/models/melgan_multiscale_discriminator.py Normal file
View File

@ -0,0 +1,41 @@
from torch import nn
from TTS.vocoder.models.melgan_discriminator import MelganDiscriminator
class MelganMultiscaleDiscriminator(nn.Module):
def __init__(self,
in_channels=1,
out_channels=1,
num_scales=3,
kernel_sizes=(5, 3),
base_channels=16,
max_channels=1024,
downsample_factors=(4, 4, 4),
pooling_kernel_size=4,
pooling_stride=2,
pooling_padding=1):
super(MelganMultiscaleDiscriminator, self).__init__()
self.discriminators = nn.ModuleList([
MelganDiscriminator(in_channels=in_channels,
out_channels=out_channels,
kernel_sizes=kernel_sizes,
base_channels=base_channels,
max_channels=max_channels,
downsample_factors=downsample_factors)
for _ in range(num_scales)
])
self.pooling = nn.AvgPool1d(kernel_size=pooling_kernel_size, stride=pooling_stride, padding=pooling_padding, count_include_pad=False)
def forward(self, x):
scores = list()
feats = list()
for disc in self.discriminators:
score, feat = disc(x)
scores.append(score)
feats.append(feat)
x = self.pooling(x)
return scores, feats

39
vocoder/models/multiband_melgan_generator.py Normal file
View File

@ -0,0 +1,39 @@
import torch
from TTS.vocoder.models.melgan_generator import MelganGenerator
from TTS.vocoder.layers.pqmf import PQMF
class MultibandMelganGenerator(MelganGenerator):
def __init__(self,
in_channels=80,
out_channels=4,
proj_kernel=7,
base_channels=384,
upsample_factors=(2, 8, 2, 2),
res_kernel=3,
num_res_blocks=3):
super(MultibandMelganGenerator,
self).__init__(in_channels=in_channels,
out_channels=out_channels,
proj_kernel=proj_kernel,
base_channels=base_channels,
upsample_factors=upsample_factors,
res_kernel=res_kernel,
num_res_blocks=num_res_blocks)
self.pqmf_layer = PQMF(N=4, taps=62, cutoff=0.15, beta=9.0)
def pqmf_analysis(self, x):
return self.pqmf_layer.analysis(x)
def pqmf_synthesis(self, x):
return self.pqmf_layer.synthesis(x)
@torch.no_grad()
def inference(self, cond_features):
cond_features = cond_features.to(self.layers[1].weight.device)
cond_features = torch.nn.functional.pad(
cond_features,
(self.inference_padding, self.inference_padding),
'replicate')
return self.pqmf_synthesis(self.layers(cond_features))

225
vocoder/models/random_window_discriminator.py Normal file
View File

@ -0,0 +1,225 @@
import numpy as np
from torch import nn
class GBlock(nn.Module):
def __init__(self, in_channels, cond_channels, downsample_factor):
super(GBlock, self).__init__()
self.in_channels = in_channels
self.cond_channels = cond_channels
self.downsample_factor = downsample_factor
self.start = nn.Sequential(
nn.AvgPool1d(downsample_factor, stride=downsample_factor),
nn.ReLU(),
nn.Conv1d(in_channels, in_channels * 2, kernel_size=3, padding=1))
self.lc_conv1d = nn.Conv1d(cond_channels,
in_channels * 2,
kernel_size=1)
self.end = nn.Sequential(
nn.ReLU(),
nn.Conv1d(in_channels * 2,
in_channels * 2,
kernel_size=3,
dilation=2,
padding=2))
self.residual = nn.Sequential(
nn.Conv1d(in_channels, in_channels * 2, kernel_size=1),
nn.AvgPool1d(downsample_factor, stride=downsample_factor))
def forward(self, inputs, conditions):
outputs = self.start(inputs) + self.lc_conv1d(conditions)
outputs = self.end(outputs)
residual_outputs = self.residual(inputs)
outputs = outputs + residual_outputs
return outputs
class DBlock(nn.Module):
def __init__(self, in_channels, out_channels, downsample_factor):
super(DBlock, self).__init__()
self.in_channels = in_channels
self.downsample_factor = downsample_factor
self.out_channels = out_channels
self.donwsample_layer = nn.AvgPool1d(downsample_factor,
stride=downsample_factor)
self.layers = nn.Sequential(
nn.ReLU(),
nn.Conv1d(in_channels, out_channels, kernel_size=3, padding=1),
nn.ReLU(),
nn.Conv1d(out_channels,
out_channels,
kernel_size=3,
dilation=2,
padding=2))
self.residual = nn.Sequential(
nn.Conv1d(in_channels, out_channels, kernel_size=1), )
def forward(self, inputs):
if self.downsample_factor > 1:
outputs = self.layers(self.donwsample_layer(inputs))\
+ self.donwsample_layer(self.residual(inputs))
else:
outputs = self.layers(inputs) + self.residual(inputs)
return outputs
class ConditionalDiscriminator(nn.Module):
def __init__(self,
in_channels,
cond_channels,
downsample_factors=(2, 2, 2),
out_channels=(128, 256)):
super(ConditionalDiscriminator, self).__init__()
assert len(downsample_factors) == len(out_channels) + 1
self.in_channels = in_channels
self.cond_channels = cond_channels
self.downsample_factors = downsample_factors
self.out_channels = out_channels
self.pre_cond_layers = nn.ModuleList()
self.post_cond_layers = nn.ModuleList()
# layers before condition features
self.pre_cond_layers += [DBlock(in_channels, 64, 1)]
in_channels = 64
for (i, channel) in enumerate(out_channels):
self.pre_cond_layers.append(
DBlock(in_channels, channel, downsample_factors[i]))
in_channels = channel
# condition block
self.cond_block = GBlock(in_channels, cond_channels,
downsample_factors[-1])
# layers after condition block
self.post_cond_layers += [
DBlock(in_channels * 2, in_channels * 2, 1),
DBlock(in_channels * 2, in_channels * 2, 1),
nn.AdaptiveAvgPool1d(1),
nn.Conv1d(in_channels * 2, 1, kernel_size=1),
]
def forward(self, inputs, conditions):
batch_size = inputs.size()[0]
outputs = inputs.view(batch_size, self.in_channels, -1)
for layer in self.pre_cond_layers:
outputs = layer(outputs)
outputs = self.cond_block(outputs, conditions)
for layer in self.post_cond_layers:
outputs = layer(outputs)
return outputs
class UnconditionalDiscriminator(nn.Module):
def __init__(self,
in_channels,
base_channels=64,
downsample_factors=(8, 4),
out_channels=(128, 256)):
super(UnconditionalDiscriminator, self).__init__()
self.downsample_factors = downsample_factors
self.in_channels = in_channels
self.downsample_factors = downsample_factors
self.out_channels = out_channels
self.layers = nn.ModuleList()
self.layers += [DBlock(self.in_channels, base_channels, 1)]
in_channels = base_channels
for (i, factor) in enumerate(downsample_factors):
self.layers.append(DBlock(in_channels, out_channels[i], factor))
in_channels *= 2
self.layers += [
DBlock(in_channels, in_channels, 1),
DBlock(in_channels, in_channels, 1),
nn.AdaptiveAvgPool1d(1),
nn.Conv1d(in_channels, 1, kernel_size=1),
]
def forward(self, inputs):
batch_size = inputs.size()[0]
outputs = inputs.view(batch_size, self.in_channels, -1)
for layer in self.layers:
outputs = layer(outputs)
return outputs
class RandomWindowDiscriminator(nn.Module):
"""Random Window Discriminator as described in
http://arxiv.org/abs/1909.11646"""
def __init__(self,
cond_channels,
hop_length,
uncond_disc_donwsample_factors=(8, 4),
cond_disc_downsample_factors=((8, 4, 2, 2, 2), (8, 4, 2, 2),
(8, 4, 2), (8, 4), (4, 2, 2)),
cond_disc_out_channels=((128, 128, 256, 256), (128, 256, 256),
(128, 256), (256, ), (128, 256)),
window_sizes=(512, 1024, 2048, 4096, 8192)):
super(RandomWindowDiscriminator, self).__init__()
self.cond_channels = cond_channels
self.window_sizes = window_sizes
self.hop_length = hop_length
self.base_window_size = self.hop_length * 2
self.ks = [ws // self.base_window_size for ws in window_sizes]
# check arguments
assert len(cond_disc_downsample_factors) == len(
cond_disc_out_channels) == len(window_sizes)
for ws in window_sizes:
assert ws % hop_length == 0
for idx, cf in enumerate(cond_disc_downsample_factors):
assert np.prod(cf) == hop_length // self.ks[idx]
# define layers
self.unconditional_discriminators = nn.ModuleList([])
for k in self.ks:
layer = UnconditionalDiscriminator(
in_channels=k,
base_channels=64,
downsample_factors=uncond_disc_donwsample_factors)
self.unconditional_discriminators.append(layer)
self.conditional_discriminators = nn.ModuleList([])
for idx, k in enumerate(self.ks):
layer = ConditionalDiscriminator(
in_channels=k,
cond_channels=cond_channels,
downsample_factors=cond_disc_downsample_factors[idx],
out_channels=cond_disc_out_channels[idx])
self.conditional_discriminators.append(layer)
def forward(self, x, c):
scores = []
feats = []
# unconditional pass
for (window_size, layer) in zip(self.window_sizes,
self.unconditional_discriminators):
index = np.random.randint(x.shape[-1] - window_size)
score = layer(x[:, :, index:index + window_size])
scores.append(score)
# conditional pass
for (window_size, layer) in zip(self.window_sizes,
self.conditional_discriminators):
frame_size = window_size // self.hop_length
lc_index = np.random.randint(c.shape[-1] - frame_size)
sample_index = lc_index * self.hop_length
x_sub = x[:, :,
sample_index:(lc_index + frame_size) * self.hop_length]
c_sub = c[:, :, lc_index:lc_index + frame_size]
score = layer(x_sub, c_sub)
scores.append(score)
return scores, feats

678
vocoder/notebooks/Untitled.ipynb Normal file
View File

File diff suppressed because one or more lines are too long

6
vocoder/notebooks/Untitled1.ipynb Normal file
View File

@ -0,0 +1,6 @@
{
"cells": [],
"metadata": {},
"nbformat": 4,
"nbformat_minor": 4
}

BIN
vocoder/pqmf_output.wav Normal file
View File

Binary file not shown.

1
vocoder/tests/__init__.py Normal file
View File

@ -0,0 +1 @@

24
vocoder/tests/test_config.json Normal file
View File

@ -0,0 +1,24 @@
{
"audio":{
"num_mels": 80, // size of the mel spec frame.
"num_freq": 513, // number of stft frequency levels. Size of the linear spectogram frame.
"sample_rate": 22050, // wav sample-rate. If different than the original data, it is resampled.
"frame_length_ms": null, // stft window length in ms.
"frame_shift_ms": null, // stft window hop-lengh in ms.
"hop_length": 256,
"win_length": 1024,
"preemphasis": 0.97, // pre-emphasis to reduce spec noise and make it more structured. If 0.0, no -pre-emphasis.
"min_level_db": -100, // normalization range
"ref_level_db": 20, // reference level db, theoretically 20db is the sound of air.
"power": 1.5, // value to sharpen wav signals after GL algorithm.
"griffin_lim_iters": 30,// #griffin-lim iterations. 30-60 is a good range. Larger the value, slower the generation.
"signal_norm": true, // normalize the spec values in range [0, 1]
"symmetric_norm": true, // move normalization to range [-1, 1]
"clip_norm": true, // clip normalized values into the range.
"max_norm": 4, // scale normalization to range [-max_norm, max_norm] or [0, max_norm]
"mel_fmin": 0, // minimum freq level for mel-spec. ~50 for male and ~95 for female voices. Tune for dataset!!
"mel_fmax": 8000, // maximum freq level for mel-spec. Tune for dataset!!
"do_trim_silence": false
}
}

95
vocoder/tests/test_datasets.py Normal file
View File

@ -0,0 +1,95 @@
import os
import numpy as np
from torch.utils.data import DataLoader
from TTS.vocoder.datasets.gan_dataset import GANDataset
from TTS.vocoder.datasets.preprocess import load_wav_data
from TTS.utils.audio import AudioProcessor
from TTS.utils.io import load_config
file_path = os.path.dirname(os.path.realpath(__file__))
OUTPATH = os.path.join(file_path, "../../tests/outputs/loader_tests/")
os.makedirs(OUTPATH, exist_ok=True)
C = load_config(os.path.join(file_path, 'test_config.json'))
test_data_path = os.path.join(file_path, "../../tests/data/ljspeech/")
ok_ljspeech = os.path.exists(test_data_path)
def gan_dataset_case(batch_size, seq_len, hop_len, conv_pad, return_segments, use_noise_augment, use_cache, num_workers):
''' run dataloader with given parameters and check conditions '''
ap = AudioProcessor(**C.audio)
_, train_items = load_wav_data(test_data_path, 10)
dataset = GANDataset(ap,
train_items,
seq_len=seq_len,
hop_len=hop_len,
pad_short=2000,
conv_pad=conv_pad,
return_segments=return_segments,
use_noise_augment=use_noise_augment,
use_cache=use_cache)
loader = DataLoader(dataset=dataset,
batch_size=batch_size,
shuffle=True,
num_workers=num_workers,
pin_memory=True,
drop_last=True)
max_iter = 10
count_iter = 0
# return random segments or return the whole audio
if return_segments:
for item1, _ in loader:
feat1, wav1 = item1
# feat2, wav2 = item2
expected_feat_shape = (batch_size, ap.num_mels, seq_len // hop_len + conv_pad * 2)
# check shapes
assert np.all(feat1.shape == expected_feat_shape), f" [!] {feat1.shape} vs {expected_feat_shape}"
assert (feat1.shape[2] - conv_pad * 2) * hop_len == wav1.shape[2]
# check feature vs audio match
if not use_noise_augment:
for idx in range(batch_size):
audio = wav1[idx].squeeze()
feat = feat1[idx]
mel = ap.melspectrogram(audio)
# the first 2 and the last frame is skipped due to the padding
# applied in spec. computation.
assert (feat - mel[:, :feat1.shape[-1]])[:, 2:-1].sum() == 0, f' [!] {(feat - mel[:, :feat1.shape[-1]])[:, 2:-1].sum()}'
count_iter += 1
# if count_iter == max_iter:
# break
else:
for item in loader:
feat, wav = item
expected_feat_shape = (batch_size, ap.num_mels, (wav.shape[-1] // hop_len) + (conv_pad * 2))
assert np.all(feat.shape == expected_feat_shape), f" [!] {feat.shape} vs {expected_feat_shape}"
assert (feat.shape[2] - conv_pad * 2) * hop_len == wav.shape[2]
count_iter += 1
if count_iter == max_iter:
break
def test_parametrized_gan_dataset():
''' test dataloader with different parameters '''
params = [
[32, C.audio['hop_length'] * 10, C.audio['hop_length'], 0, True, False, True, 0],
[32, C.audio['hop_length'] * 10, C.audio['hop_length'], 0, True, False, True, 4],
[1, C.audio['hop_length'] * 10, C.audio['hop_length'], 0, True, True, True, 0],
[1, C.audio['hop_length'], C.audio['hop_length'], 0, True, True, True, 0],
[1, C.audio['hop_length'] * 10, C.audio['hop_length'], 2, True, True, True, 0],
[1, C.audio['hop_length'] * 10, C.audio['hop_length'], 0, False, True, True, 0],
[1, C.audio['hop_length'] * 10, C.audio['hop_length'], 0, True, False, True, 0],
[1, C.audio['hop_length'] * 10, C.audio['hop_length'], 0, True, True, False, 0],
[1, C.audio['hop_length'] * 10, C.audio['hop_length'], 0, False, False, False, 0],
]
for param in params:
print(param)
gan_dataset_case(*param)

61
vocoder/tests/test_losses.py Normal file
View File

@ -0,0 +1,61 @@
import os
import torch
from TTS.vocoder.layers.losses import TorchSTFT, STFTLoss, MultiScaleSTFTLoss
from TTS.tests import get_tests_path, get_tests_input_path, get_tests_output_path
from TTS.utils.audio import AudioProcessor
from TTS.utils.io import load_config
TESTS_PATH = get_tests_path()
OUT_PATH = os.path.join(get_tests_output_path(), "audio_tests")
os.makedirs(OUT_PATH, exist_ok=True)
WAV_FILE = os.path.join(get_tests_input_path(), "example_1.wav")
file_path = os.path.dirname(os.path.realpath(__file__))
C = load_config(os.path.join(file_path, 'test_config.json'))
ap = AudioProcessor(**C.audio)
def test_torch_stft():
torch_stft = TorchSTFT(ap.n_fft, ap.hop_length, ap.win_length)
# librosa stft
wav = ap.load_wav(WAV_FILE)
M_librosa = abs(ap._stft(wav)) # pylint: disable=protected-access
# torch stft
wav = torch.from_numpy(wav[None, :]).float()
M_torch = torch_stft(wav)
# check the difference b/w librosa and torch outputs
assert (M_librosa - M_torch[0].data.numpy()).max() < 1e-5
def test_stft_loss():
stft_loss = STFTLoss(ap.n_fft, ap.hop_length, ap.win_length)
wav = ap.load_wav(WAV_FILE)
wav = torch.from_numpy(wav[None, :]).float()
loss_m, loss_sc = stft_loss(wav, wav)
assert loss_m + loss_sc == 0
loss_m, loss_sc = stft_loss(wav, torch.rand_like(wav))
assert loss_sc < 1.0
assert loss_m + loss_sc > 0
def test_multiscale_stft_loss():
stft_loss = MultiScaleSTFTLoss([ap.n_fft//2, ap.n_fft, ap.n_fft*2],
[ap.hop_length // 2, ap.hop_length, ap.hop_length * 2],
[ap.win_length // 2, ap.win_length, ap.win_length * 2])
wav = ap.load_wav(WAV_FILE)
wav = torch.from_numpy(wav[None, :]).float()
loss_m, loss_sc = stft_loss(wav, wav)
assert loss_m + loss_sc == 0
loss_m, loss_sc = stft_loss(wav, torch.rand_like(wav))
assert loss_sc < 1.0
assert loss_m + loss_sc > 0

26
vocoder/tests/test_melgan_discriminator.py Normal file
View File

@ -0,0 +1,26 @@
import numpy as np
import torch
from TTS.vocoder.models.melgan_discriminator import MelganDiscriminator
from TTS.vocoder.models.melgan_multiscale_discriminator import MelganMultiscaleDiscriminator
def test_melgan_discriminator():
model = MelganDiscriminator()
print(model)
dummy_input = torch.rand((4, 1, 256 * 10))
output, _ = model(dummy_input)
assert np.all(output.shape == (4, 1, 10))
def test_melgan_multi_scale_discriminator():
model = MelganMultiscaleDiscriminator()
print(model)
dummy_input = torch.rand((4, 1, 256 * 16))
scores, feats = model(dummy_input)
assert len(scores) == 3
assert len(scores) == len(feats)
assert np.all(scores[0].shape == (4, 1, 64))
assert np.all(feats[0][0].shape == (4, 16, 4096))
assert np.all(feats[0][1].shape == (4, 64, 1024))
assert np.all(feats[0][2].shape == (4, 256, 256))

14
vocoder/tests/test_melgan_generator.py Normal file
View File

@ -0,0 +1,14 @@
import numpy as np
import torch
from TTS.vocoder.models.melgan_generator import MelganGenerator
def test_melgan_generator():
model = MelganGenerator()
print(model)
dummy_input = torch.rand((4, 80, 64))
output = model(dummy_input)
assert np.all(output.shape == (4, 1, 64 * 256))
output = model.inference(dummy_input)
assert np.all(output.shape == (4, 1, (64 + 4) * 256))

28
vocoder/tests/test_pqmf.py Normal file
View File

@ -0,0 +1,28 @@
import os
import torch
import soundfile as sf
from librosa.core import load
from TTS.tests import get_tests_path, get_tests_input_path
from TTS.vocoder.layers.pqmf import PQMF
TESTS_PATH = get_tests_path()
WAV_FILE = os.path.join(get_tests_input_path(), "example_1.wav")
def test_pqmf():
w, sr = load(WAV_FILE)
layer = PQMF(N=4, taps=62, cutoff=0.15, beta=9.0)
w, sr = load(WAV_FILE)
w2 = torch.from_numpy(w[None, None, :])
b2 = layer.analysis(w2)
w2_ = layer.synthesis(b2)
print(w2_.max())
print(w2_.min())
print(w2_.mean())
sf.write('pqmf_output.wav', w2_.flatten().detach(), sr)

21
vocoder/tests/test_rwd.py Normal file
View File

@ -0,0 +1,21 @@
import torch
import numpy as np
from TTS.vocoder.models.random_window_discriminator import RandomWindowDiscriminator
def test_rwd():
layer = RandomWindowDiscriminator(cond_channels=80,
window_sizes=(512, 1024, 2048, 4096,
8192),
cond_disc_downsample_factors=[
(8, 4, 2, 2, 2), (8, 4, 2, 2),
(8, 4, 2), (8, 4), (4, 2, 2)
],
hop_length=256)
x = torch.rand([4, 1, 22050])
c = torch.rand([4, 80, 22050 // 256])
scores, _ = layer(x, c)
assert len(scores) == 10
assert np.all(scores[0].shape == (4, 1, 1))

112
vocoder/tf/convert_melgan_torch_to_tf.py Normal file
View File

@ -0,0 +1,112 @@
import argparse
import os
import numpy as np
import tensorflow as tf
import torch
from fuzzywuzzy import fuzz
from TTS.utils.io import load_config
from TTS.vocoder.tf.utils.convert_torch_to_tf_utils import (
compare_torch_tf, convert_tf_name, transfer_weights_torch_to_tf)
from TTS.vocoder.tf.utils.generic_utils import \
setup_generator as setup_tf_generator
from TTS.vocoder.tf.utils.io import save_checkpoint
from TTS.vocoder.utils.generic_utils import setup_generator
# prevent GPU use
os.environ['CUDA_VISIBLE_DEVICES'] = ''
# define args
parser = argparse.ArgumentParser()
parser.add_argument('--torch_model_path',
type=str,
help='Path to target torch model to be converted to TF.')
parser.add_argument('--config_path',
type=str,
help='Path to config file of torch model.')
parser.add_argument(
'--output_path',
type=str,
help='path to output file including file name to save TF model.')
args = parser.parse_args()
# load model config
config_path = args.config_path
c = load_config(config_path)
num_speakers = 0
# init torch model
model = setup_generator(c)
checkpoint = torch.load(args.torch_model_path,
map_location=torch.device('cpu'))
state_dict = checkpoint['model']
model.load_state_dict(state_dict)
model.remove_weight_norm()
state_dict = model.state_dict()
# init tf model
model_tf = setup_tf_generator(c)
common_sufix = '/.ATTRIBUTES/VARIABLE_VALUE'
# get tf_model graph by passing an input
# B x D x T
dummy_input = tf.random.uniform((7, 80, 64), dtype=tf.float32)
mel_pred = model_tf(dummy_input, training=False)
# get tf variables
tf_vars = model_tf.weights
# match variable names with fuzzy logic
torch_var_names = list(state_dict.keys())
tf_var_names = [we.name for we in model_tf.weights]
var_map = []
for tf_name in tf_var_names:
# skip re-mapped layer names
if tf_name in [name[0] for name in var_map]:
continue
tf_name_edited = convert_tf_name(tf_name)
ratios = [
fuzz.ratio(torch_name, tf_name_edited)
for torch_name in torch_var_names
]
max_idx = np.argmax(ratios)
matching_name = torch_var_names[max_idx]
del torch_var_names[max_idx]
var_map.append((tf_name, matching_name))
# pass weights
tf_vars = transfer_weights_torch_to_tf(tf_vars, dict(var_map), state_dict)
# Compare TF and TORCH models
# check embedding outputs
model.eval()
dummy_input_torch = torch.ones((1, 80, 10))
dummy_input_tf = tf.convert_to_tensor(dummy_input_torch.numpy())
dummy_input_tf = tf.transpose(dummy_input_tf, perm=[0, 2, 1])
dummy_input_tf = tf.expand_dims(dummy_input_tf, 2)
out_torch = model.layers[0](dummy_input_torch)
out_tf = model_tf.model_layers[0](dummy_input_tf)
out_tf_ = tf.transpose(out_tf, perm=[0, 3, 2, 1])[:, :, 0, :]
assert compare_torch_tf(out_torch, out_tf_) < 1e-5
for i in range(1, len(model.layers)):
print(f"{i} -> {model.layers[i]} vs {model_tf.model_layers[i]}")
out_torch = model.layers[i](out_torch)
out_tf = model_tf.model_layers[i](out_tf)
out_tf_ = tf.transpose(out_tf, perm=[0, 3, 2, 1])[:, :, 0, :]
diff = compare_torch_tf(out_torch, out_tf_)
assert diff < 1e-5, diff
dummy_input_torch = torch.ones((1, 80, 10))
dummy_input_tf = tf.convert_to_tensor(dummy_input_torch.numpy())
output_torch = model.inference(dummy_input_torch)
output_tf = model_tf(dummy_input_tf, training=False)
assert compare_torch_tf(output_torch, output_tf) < 1e-5, compare_torch_tf(
output_torch, output_tf)
# save tf model
save_checkpoint(model_tf, checkpoint['step'], checkpoint['epoch'],
args.output_path)
print(' > Model conversion is successfully completed :).')

58
vocoder/tf/layers/melgan.py Normal file
View File

@ -0,0 +1,58 @@
import tensorflow as tf
class ReflectionPad1d(tf.keras.layers.Layer):
def __init__(self, padding):
super(ReflectionPad1d, self).__init__()
self.padding = padding
def call(self, x):
print(x.shape)
return tf.pad(x, [[0, 0], [self.padding, self.padding], [0, 0], [0, 0]], "REFLECT")
class ResidualStack(tf.keras.layers.Layer):
def __init__(self, channels, num_res_blocks, kernel_size, name):
super(ResidualStack, self).__init__(name=name)
assert (kernel_size - 1) % 2 == 0, " [!] kernel_size has to be odd."
base_padding = (kernel_size - 1) // 2
self.blocks = []
num_layers = 2
for idx in range(num_res_blocks):
layer_kernel_size = kernel_size
layer_dilation = layer_kernel_size**idx
layer_padding = base_padding * layer_dilation
block = [
tf.keras.layers.LeakyReLU(0.2),
ReflectionPad1d(layer_padding),
tf.keras.layers.Conv2D(filters=channels,
kernel_size=(kernel_size, 1),
dilation_rate=(layer_dilation, 1),
use_bias=True,
padding='valid',
name=f'blocks.{idx}.{num_layers}'),
tf.keras.layers.LeakyReLU(0.2),
tf.keras.layers.Conv2D(filters=channels,
kernel_size=(1, 1),
use_bias=True,
name=f'blocks.{idx}.{num_layers + 2}')
]
self.blocks.append(block)
self.shortcuts = [
tf.keras.layers.Conv2D(channels,
kernel_size=1,
use_bias=True,
name=f'shortcuts.{i}')
for i in range(num_res_blocks)
]
def call(self, x):
# breakpoint()
for block, shortcut in zip(self.blocks, self.shortcuts):
res = shortcut(x)
for layer in block:
x = layer(x)
x += res
return x

66
vocoder/tf/layers/pqmf.py Normal file
View File

@ -0,0 +1,66 @@
import numpy as np
import tensorflow as tf
from scipy import signal as sig
class PQMF(tf.keras.layers.Layer):
def __init__(self, N=4, taps=62, cutoff=0.15, beta=9.0):
super(PQMF, self).__init__()
# define filter coefficient
self.N = N
self.taps = taps
self.cutoff = cutoff
self.beta = beta
QMF = sig.firwin(taps + 1, cutoff, window=('kaiser', beta))
H = np.zeros((N, len(QMF)))
G = np.zeros((N, len(QMF)))
for k in range(N):
constant_factor = (2 * k + 1) * (np.pi /
(2 * N)) * (np.arange(taps + 1) -
((taps - 1) / 2))
phase = (-1)**k * np.pi / 4
H[k] = 2 * QMF * np.cos(constant_factor + phase)
G[k] = 2 * QMF * np.cos(constant_factor - phase)
# [N, 1, taps + 1] == [filter_width, in_channels, out_channels]
self.H = np.transpose(H[:, None, :], (2, 1, 0)).astype('float32')
self.G = np.transpose(G[None, :, :], (2, 1, 0)).astype('float32')
# filter for downsampling & upsampling
updown_filter = np.zeros((N, N, N), dtype=np.float32)
for k in range(N):
updown_filter[0, k, k] = 1.0
self.updown_filter = updown_filter.astype(np.float32)
def analysis(self, x):
"""
x : B x 1 x T
"""
x = tf.transpose(x, perm=[0, 2, 1])
x = tf.pad(x, [[0, 0], [self.taps // 2, self.taps // 2], [0, 0]], constant_values=0.0)
x = tf.nn.conv1d(x, self.H, stride=1, padding='VALID')
x = tf.nn.conv1d(x,
self.updown_filter,
stride=self.N,
padding='VALID')
x = tf.transpose(x, perm=[0, 2, 1])
return x
def synthesis(self, x):
"""
x : B x 1 x T
"""
x = tf.transpose(x, perm=[0, 2, 1])
x = tf.nn.conv1d_transpose(
x,
self.updown_filter * self.N,
strides=self.N,
output_shape=(tf.shape(x)[0], tf.shape(x)[1] * self.N,
self.N))
x = tf.pad(x, [[0, 0], [self.taps // 2, self.taps // 2], [0, 0]], constant_values=0.0)
x = tf.nn.conv1d(x, self.G, stride=1, padding="VALID")
x = tf.transpose(x, perm=[0, 2, 1])
return x

103
vocoder/tf/models/melgan_generator.py Normal file
View File

@ -0,0 +1,103 @@
import logging
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3' # FATAL
logging.getLogger('tensorflow').setLevel(logging.FATAL)
import tensorflow as tf
from TTS.vocoder.tf.layers.melgan import ResidualStack, ReflectionPad1d
class MelganGenerator(tf.keras.models.Model): # pylint: disable=too-many-ancestors
""" Melgan Generator TF implementation dedicated for inference with no
weight norm """
def __init__(self,
in_channels=80,
out_channels=1,
proj_kernel=7,
base_channels=512,
upsample_factors=(8, 8, 2, 2),
res_kernel=3,
num_res_blocks=3):
super(MelganGenerator, self).__init__()
self.in_channels = in_channels
# assert model parameters
assert (proj_kernel -
1) % 2 == 0, " [!] proj_kernel should be an odd number."
# setup additional model parameters
base_padding = (proj_kernel - 1) // 2
act_slope = 0.2
self.inference_padding = 2
# initial layer
self.initial_layer = [
ReflectionPad1d(base_padding),
tf.keras.layers.Conv2D(filters=base_channels,
kernel_size=(proj_kernel, 1),
strides=1,
padding='valid',
use_bias=True,
name="1")
]
num_layers = 3 # count number of layers for layer naming
# upsampling layers and residual stacks
self.upsample_layers = []
for idx, upsample_factor in enumerate(upsample_factors):
layer_out_channels = base_channels // (2**(idx + 1))
layer_filter_size = upsample_factor * 2
layer_stride = upsample_factor
# layer_output_padding = upsample_factor % 2
self.upsample_layers += [
tf.keras.layers.LeakyReLU(act_slope),
tf.keras.layers.Conv2DTranspose(
filters=layer_out_channels,
kernel_size=(layer_filter_size, 1),
strides=(layer_stride, 1),
padding='same',
# output_padding=layer_output_padding,
use_bias=True,
name=f'{num_layers}'),
ResidualStack(channels=layer_out_channels,
num_res_blocks=num_res_blocks,
kernel_size=res_kernel,
name=f'layers.{num_layers + 1}')
]
num_layers += num_res_blocks - 1
self.upsample_layers += [tf.keras.layers.LeakyReLU(act_slope)]
# final layer
self.final_layers = [
ReflectionPad1d(base_padding),
tf.keras.layers.Conv2D(filters=out_channels,
kernel_size=(proj_kernel, 1),
use_bias=True,
name=f'layers.{num_layers + 1}'),
tf.keras.layers.Activation("tanh")
]
# self.model_layers = tf.keras.models.Sequential(self.initial_layer + self.upsample_layers + self.final_layers, name="layers")
self.model_layers = self.initial_layer + self.upsample_layers + self.final_layers
def call(self, c, training=False):
"""
c : B x C x T
"""
if training:
raise NotImplementedError()
return self.inference(c)
def inference(self, c):
c = tf.transpose(c, perm=[0, 2, 1])
c = tf.expand_dims(c, 2)
c = tf.pad(c, [[0, 0], [self.inference_padding, self.inference_padding], [0, 0], [0, 0]], "REFLECT")
o = c
for layer in self.model_layers:
o = layer(o)
# o = self.model_layers(c)
o = tf.transpose(o, perm=[0, 3, 2, 1])
return o[:, :, 0, :]

46
vocoder/tf/models/multiband_melgan_generator.py Normal file
View File

@ -0,0 +1,46 @@
import tensorflow as tf
from TTS.vocoder.tf.models.melgan_generator import MelganGenerator
from TTS.vocoder.tf.layers.pqmf import PQMF
class MultibandMelganGenerator(MelganGenerator): # pylint: disable=too-many-ancestors
def __init__(self,
in_channels=80,
out_channels=4,
proj_kernel=7,
base_channels=384,
upsample_factors=(2, 8, 2, 2),
res_kernel=3,
num_res_blocks=3):
super(MultibandMelganGenerator,
self).__init__(in_channels=in_channels,
out_channels=out_channels,
proj_kernel=proj_kernel,
base_channels=base_channels,
upsample_factors=upsample_factors,
res_kernel=res_kernel,
num_res_blocks=num_res_blocks)
self.pqmf_layer = PQMF(N=4, taps=62, cutoff=0.15, beta=9.0)
def pqmf_analysis(self, x):
return self.pqmf_layer.analysis(x)
def pqmf_synthesis(self, x):
return self.pqmf_layer.synthesis(x)
# def call(self, c, training=False):
# if training:
# raise NotImplementedError()
# return self.inference(c)
def inference(self, c):
c = tf.transpose(c, perm=[0, 2, 1])
c = tf.expand_dims(c, 2)
c = tf.pad(c, [[0, 0], [self.inference_padding, self.inference_padding], [0, 0], [0, 0]], "REFLECT")
o = c
for layer in self.model_layers:
o = layer(o)
o = tf.transpose(o, perm=[0, 3, 2, 1])
o = self.pqmf_layer.synthesis(o[:, :, 0, :])
return o

0
vocoder/tf/utils/__init__.py Normal file
View File

45
vocoder/tf/utils/convert_torch_to_tf_utils.py Normal file
View File

@ -0,0 +1,45 @@
import numpy as np
import tensorflow as tf
def compare_torch_tf(torch_tensor, tf_tensor):
""" Compute the average absolute difference b/w torch and tf tensors """
return abs(torch_tensor.detach().numpy() - tf_tensor.numpy()).mean()
def convert_tf_name(tf_name):
""" Convert certain patterns in TF layer names to Torch patterns """
tf_name_tmp = tf_name
tf_name_tmp = tf_name_tmp.replace(':0', '')
tf_name_tmp = tf_name_tmp.replace('/forward_lstm/lstm_cell_1/recurrent_kernel', '/weight_hh_l0')
tf_name_tmp = tf_name_tmp.replace('/forward_lstm/lstm_cell_2/kernel', '/weight_ih_l1')
tf_name_tmp = tf_name_tmp.replace('/recurrent_kernel', '/weight_hh')
tf_name_tmp = tf_name_tmp.replace('/kernel', '/weight')
tf_name_tmp = tf_name_tmp.replace('/gamma', '/weight')
tf_name_tmp = tf_name_tmp.replace('/beta', '/bias')
tf_name_tmp = tf_name_tmp.replace('/', '.')
return tf_name_tmp
def transfer_weights_torch_to_tf(tf_vars, var_map_dict, state_dict):
""" Transfer weigths from torch state_dict to TF variables """
print(" > Passing weights from Torch to TF ...")
for tf_var in tf_vars:
torch_var_name = var_map_dict[tf_var.name]
print(f' | > {tf_var.name} <-- {torch_var_name}')
# if tuple, it is a bias variable
if 'kernel' in tf_var.name:
torch_weight = state_dict[torch_var_name]
numpy_weight = torch_weight.permute([2, 1, 0]).numpy()[:, None, :, :]
if 'bias' in tf_var.name:
torch_weight = state_dict[torch_var_name]
numpy_weight = torch_weight
assert np.all(tf_var.shape == numpy_weight.shape), f" [!] weight shapes does not match: {tf_var.name} vs {torch_var_name} --> {tf_var.shape} vs {numpy_weight.shape}"
tf.keras.backend.set_value(tf_var, numpy_weight)
return tf_vars
def load_tf_vars(model_tf, tf_vars):
for tf_var in tf_vars:
model_tf.get_layer(tf_var.name).set_weights(tf_var)
return model_tf

35
vocoder/tf/utils/generic_utils.py Normal file
View File

@ -0,0 +1,35 @@
import re
import importlib
def to_camel(text):
text = text.capitalize()
return re.sub(r'(?!^)_([a-zA-Z])', lambda m: m.group(1).upper(), text)
def setup_generator(c):
print(" > Generator Model: {}".format(c.generator_model))
MyModel = importlib.import_module('TTS.vocoder.tf.models.' +
c.generator_model.lower())
MyModel = getattr(MyModel, to_camel(c.generator_model))
if c.generator_model in 'melgan_generator':
model = MyModel(
in_channels=c.audio['num_mels'],
out_channels=1,
proj_kernel=7,
base_channels=512,
upsample_factors=c.generator_model_params['upsample_factors'],
res_kernel=3,
num_res_blocks=c.generator_model_params['num_res_blocks'])
if c.generator_model in 'melgan_fb_generator':
pass
if c.generator_model in 'multiband_melgan_generator':
model = MyModel(
in_channels=c.audio['num_mels'],
out_channels=4,
proj_kernel=7,
base_channels=384,
upsample_factors=c.generator_model_params['upsample_factors'],
res_kernel=3,
num_res_blocks=c.generator_model_params['num_res_blocks'])
return model

27
vocoder/tf/utils/io.py Normal file
View File

@ -0,0 +1,27 @@
import datetime
import pickle
import tensorflow as tf
def save_checkpoint(model, current_step, epoch, output_path, **kwargs):
""" Save TF Vocoder model """
state = {
'model': model.weights,
'step': current_step,
'epoch': epoch,
'date': datetime.date.today().strftime("%B %d, %Y"),
}
state.update(kwargs)
pickle.dump(state, open(output_path, 'wb'))
def load_checkpoint(model, checkpoint_path):
""" Load TF Vocoder model """
checkpoint = pickle.load(open(checkpoint_path, 'rb'))
chkp_var_dict = {var.name: var.numpy() for var in checkpoint['model']}
tf_vars = model.weights
for tf_var in tf_vars:
layer_name = tf_var.name
chkp_var_value = chkp_var_dict[layer_name]
tf.keras.backend.set_value(tf_var, chkp_var_value)
return model

658
vocoder/train.py Normal file
View File

@ -0,0 +1,658 @@
import argparse
import glob
import os
import sys
import time
import traceback
import torch
from torch.utils.data import DataLoader
from inspect import signature
from TTS.utils.audio import AudioProcessor
from TTS.utils.generic_utils import (KeepAverage, count_parameters,
create_experiment_folder, get_git_branch,
remove_experiment_folder, set_init_dict)
from TTS.utils.io import copy_config_file, load_config
from TTS.utils.radam import RAdam
from TTS.utils.tensorboard_logger import TensorboardLogger
from TTS.utils.training import setup_torch_training_env
from TTS.vocoder.datasets.gan_dataset import GANDataset
from TTS.vocoder.datasets.preprocess import load_wav_data, load_wav_feat_data
# from distribute import (DistributedSampler, apply_gradient_allreduce,
# init_distributed, reduce_tensor)
from TTS.vocoder.layers.losses import DiscriminatorLoss, GeneratorLoss
from TTS.vocoder.utils.io import save_checkpoint, save_best_model
from TTS.vocoder.utils.console_logger import ConsoleLogger
from TTS.vocoder.utils.generic_utils import (check_config, plot_results,
setup_discriminator,
setup_generator)
use_cuda, num_gpus = setup_torch_training_env(True, True)
def setup_loader(ap, is_val=False, verbose=False):
if is_val and not c.run_eval:
loader = None
else:
dataset = GANDataset(ap=ap,
items=eval_data if is_val else train_data,
seq_len=c.seq_len,
hop_len=ap.hop_length,
pad_short=c.pad_short,
conv_pad=c.conv_pad,
is_training=not is_val,
return_segments=not is_val,
use_noise_augment=c.use_noise_augment,
use_cache=c.use_cache,
verbose=verbose)
dataset.shuffle_mapping()
# sampler = DistributedSampler(dataset) if num_gpus > 1 else None
loader = DataLoader(dataset,
batch_size=1 if is_val else c.batch_size,
shuffle=True,
drop_last=False,
sampler=None,
num_workers=c.num_val_loader_workers
if is_val else c.num_loader_workers,
pin_memory=False)
return loader
def format_data(data):
if isinstance(data[0], list):
# setup input data
c_G, x_G = data[0]
c_D, x_D = data[1]
# dispatch data to GPU
if use_cuda:
c_G = c_G.cuda(non_blocking=True)
x_G = x_G.cuda(non_blocking=True)
c_D = c_D.cuda(non_blocking=True)
x_D = x_D.cuda(non_blocking=True)
return c_G, x_G, c_D, x_D
# return a whole audio segment
co, x = data
if use_cuda:
co = co.cuda(non_blocking=True)
x = x.cuda(non_blocking=True)
return co, x, None, None
def train(model_G, criterion_G, optimizer_G, model_D, criterion_D, optimizer_D,
scheduler_G, scheduler_D, ap, global_step, epoch):
data_loader = setup_loader(ap, is_val=False, verbose=(epoch == 0))
model_G.train()
model_D.train()
epoch_time = 0
keep_avg = KeepAverage()
if use_cuda:
batch_n_iter = int(
len(data_loader.dataset) / (c.batch_size * num_gpus))
else:
batch_n_iter = int(len(data_loader.dataset) / c.batch_size)
end_time = time.time()
c_logger.print_train_start()
for num_iter, data in enumerate(data_loader):
start_time = time.time()
# format data
c_G, y_G, c_D, y_D = format_data(data)
loader_time = time.time() - end_time
global_step += 1
##############################
# GENERATOR
##############################
# generator pass
y_hat = model_G(c_G)
y_hat_sub = None
y_G_sub = None
y_hat_vis = y_hat # for visualization
# PQMF formatting
if y_hat.shape[1] > 1:
y_hat_sub = y_hat
y_hat = model_G.pqmf_synthesis(y_hat)
y_hat_vis = y_hat
y_G_sub = model_G.pqmf_analysis(y_G)
if global_step > c.steps_to_start_discriminator:
# run D with or without cond. features
if len(signature(model_D.forward).parameters) == 2:
D_out_fake = model_D(y_hat, c_G)
else:
D_out_fake = model_D(y_hat)
D_out_real = None
if c.use_feat_match_loss:
with torch.no_grad():
D_out_real = model_D(y_G)
# format D outputs
if isinstance(D_out_fake, tuple):
scores_fake, feats_fake = D_out_fake
if D_out_real is None:
feats_real = None
else:
_, feats_real = D_out_real
else:
scores_fake = D_out_fake
else:
scores_fake, feats_fake, feats_real = None, None, None
# compute losses
loss_G_dict = criterion_G(y_hat, y_G, scores_fake, feats_fake,
feats_real, y_hat_sub, y_G_sub)
loss_G = loss_G_dict['G_loss']
# optimizer generator
optimizer_G.zero_grad()
loss_G.backward()
if c.gen_clip_grad > 0:
torch.nn.utils.clip_grad_norm_(model_G.parameters(),
c.gen_clip_grad)
optimizer_G.step()
if scheduler_G is not None:
scheduler_G.step()
loss_dict = dict()
for key, value in loss_G_dict.items():
if isinstance(value, int):
loss_dict[key] = value
else:
loss_dict[key] = value.item()
##############################
# DISCRIMINATOR
##############################
if global_step >= c.steps_to_start_discriminator:
# discriminator pass
with torch.no_grad():
y_hat = model_G(c_D)
# PQMF formatting
if y_hat.shape[1] > 1:
y_hat = model_G.pqmf_synthesis(y_hat)
# run D with or without cond. features
if len(signature(model_D.forward).parameters) == 2:
D_out_fake = model_D(y_hat.detach(), c_D)
D_out_real = model_D(y_D, c_D)
else:
D_out_fake = model_D(y_hat.detach())
D_out_real = model_D(y_D)
# format D outputs
if isinstance(D_out_fake, tuple):
scores_fake, feats_fake = D_out_fake
if D_out_real is None:
scores_real, feats_real = None, None
else:
scores_real, feats_real = D_out_real
else:
scores_fake = D_out_fake
scores_real = D_out_real
# compute losses
loss_D_dict = criterion_D(scores_fake, scores_real)
loss_D = loss_D_dict['D_loss']
# optimizer discriminator
optimizer_D.zero_grad()
loss_D.backward()
if c.disc_clip_grad > 0:
torch.nn.utils.clip_grad_norm_(model_D.parameters(),
c.disc_clip_grad)
optimizer_D.step()
if scheduler_D is not None:
scheduler_D.step()
for key, value in loss_D_dict.items():
if isinstance(value, (int, float)):
loss_dict[key] = value
else:
loss_dict[key] = value.item()
step_time = time.time() - start_time
epoch_time += step_time
# get current learning rates
current_lr_G = list(optimizer_G.param_groups)[0]['lr']
current_lr_D = list(optimizer_D.param_groups)[0]['lr']
# update avg stats
update_train_values = dict()
for key, value in loss_dict.items():
update_train_values['avg_' + key] = value
update_train_values['avg_loader_time'] = loader_time
update_train_values['avg_step_time'] = step_time
keep_avg.update_values(update_train_values)
# print training stats
if global_step % c.print_step == 0:
c_logger.print_train_step(batch_n_iter, num_iter, global_step,
step_time, loader_time, current_lr_G,
current_lr_D, loss_dict,
keep_avg.avg_values)
# plot step stats
if global_step % 10 == 0:
iter_stats = {
"lr_G": current_lr_G,
"lr_D": current_lr_D,
"step_time": step_time
}
iter_stats.update(loss_dict)
tb_logger.tb_train_iter_stats(global_step, iter_stats)
# save checkpoint
if global_step % c.save_step == 0:
if c.checkpoint:
# save model
save_checkpoint(model_G,
optimizer_G,
scheduler_G,
model_D,
optimizer_D,
scheduler_D,
global_step,
epoch,
OUT_PATH,
model_losses=loss_dict)
# compute spectrograms
figures = plot_results(y_hat_vis, y_G, ap, global_step,
'train')
tb_logger.tb_train_figures(global_step, figures)
# Sample audio
sample_voice = y_hat_vis[0].squeeze(0).detach().cpu().numpy()
tb_logger.tb_train_audios(global_step,
{'train/audio': sample_voice},
c.audio["sample_rate"])
end_time = time.time()
# print epoch stats
c_logger.print_train_epoch_end(global_step, epoch, epoch_time, keep_avg)
# Plot Training Epoch Stats
epoch_stats = {"epoch_time": epoch_time}
epoch_stats.update(keep_avg.avg_values)
tb_logger.tb_train_epoch_stats(global_step, epoch_stats)
# TODO: plot model stats
# if c.tb_model_param_stats:
# tb_logger.tb_model_weights(model, global_step)
return keep_avg.avg_values, global_step
@torch.no_grad()
def evaluate(model_G, criterion_G, model_D, criterion_D, ap, global_step, epoch):
data_loader = setup_loader(ap, is_val=True, verbose=(epoch == 0))
model_G.eval()
model_D.eval()
epoch_time = 0
keep_avg = KeepAverage()
end_time = time.time()
c_logger.print_eval_start()
for num_iter, data in enumerate(data_loader):
start_time = time.time()
# format data
c_G, y_G, _, _ = format_data(data)
loader_time = time.time() - end_time
global_step += 1
##############################
# GENERATOR
##############################
# generator pass
y_hat = model_G(c_G)
y_hat_sub = None
y_G_sub = None
# PQMF formatting
if y_hat.shape[1] > 1:
y_hat_sub = y_hat
y_hat = model_G.pqmf_synthesis(y_hat)
y_G_sub = model_G.pqmf_analysis(y_G)
if global_step > c.steps_to_start_discriminator:
if len(signature(model_D.forward).parameters) == 2:
D_out_fake = model_D(y_hat, c_G)
else:
D_out_fake = model_D(y_hat)
D_out_real = None
if c.use_feat_match_loss:
with torch.no_grad():
D_out_real = model_D(y_G)
# format D outputs
if isinstance(D_out_fake, tuple):
scores_fake, feats_fake = D_out_fake
if D_out_real is None:
feats_real = None
else:
_, feats_real = D_out_real
else:
scores_fake = D_out_fake
else:
scores_fake, feats_fake, feats_real = None, None, None
# compute losses
loss_G_dict = criterion_G(y_hat, y_G, scores_fake, feats_fake,
feats_real, y_hat_sub, y_G_sub)
loss_dict = dict()
for key, value in loss_G_dict.items():
if isinstance(value, (int, float)):
loss_dict[key] = value
else:
loss_dict[key] = value.item()
##############################
# DISCRIMINATOR
##############################
if global_step >= c.steps_to_start_discriminator:
# discriminator pass
with torch.no_grad():
y_hat = model_G(c_G)
# PQMF formatting
if y_hat.shape[1] > 1:
y_hat = model_G.pqmf_synthesis(y_hat)
# run D with or without cond. features
if len(signature(model_D.forward).parameters) == 2:
D_out_fake = model_D(y_hat.detach(), c_G)
D_out_real = model_D(y_G, c_G)
else:
D_out_fake = model_D(y_hat.detach())
D_out_real = model_D(y_G)
# format D outputs
if isinstance(D_out_fake, tuple):
scores_fake, feats_fake = D_out_fake
if D_out_real is None:
scores_real, feats_real = None, None
else:
scores_real, feats_real = D_out_real
else:
scores_fake = D_out_fake
scores_real = D_out_real
# compute losses
loss_D_dict = criterion_D(scores_fake, scores_real)
for key, value in loss_D_dict.items():
if isinstance(value, (int, float)):
loss_dict[key] = value
else:
loss_dict[key] = value.item()
step_time = time.time() - start_time
epoch_time += step_time
# update avg stats
update_eval_values = dict()
for key, value in loss_dict.items():
update_eval_values['avg_' + key] = value
update_eval_values['avg_loader_time'] = loader_time
update_eval_values['avg_step_time'] = step_time
keep_avg.update_values(update_eval_values)
# print eval stats
if c.print_eval:
c_logger.print_eval_step(num_iter, loss_dict, keep_avg.avg_values)
# compute spectrograms
figures = plot_results(y_hat, y_G, ap, global_step, 'eval')
tb_logger.tb_eval_figures(global_step, figures)
# Sample audio
sample_voice = y_hat[0].squeeze(0).detach().cpu().numpy()
tb_logger.tb_eval_audios(global_step, {'eval/audio': sample_voice},
c.audio["sample_rate"])
# synthesize a full voice
data_loader.return_segments = False
tb_logger.tb_eval_stats(global_step, keep_avg.avg_values)
return keep_avg.avg_values
# FIXME: move args definition/parsing inside of main?
def main(args): # pylint: disable=redefined-outer-name
# pylint: disable=global-variable-undefined
global train_data, eval_data
print(f" > Loading wavs from: {c.data_path}")
if c.feature_path is not None:
print(f" > Loading features from: {c.feature_path}")
eval_data, train_data = load_wav_feat_data(c.data_path, c.feature_path, c.eval_split_size)
else:
eval_data, train_data = load_wav_data(c.data_path, c.eval_split_size)
# setup audio processor
ap = AudioProcessor(**c.audio)
# DISTRUBUTED
# if num_gpus > 1:
# init_distributed(args.rank, num_gpus, args.group_id,
# c.distributed["backend"], c.distributed["url"])
# setup models
model_gen = setup_generator(c)
model_disc = setup_discriminator(c)
# setup optimizers
optimizer_gen = RAdam(model_gen.parameters(), lr=c.lr_gen, weight_decay=0)
optimizer_disc = RAdam(model_disc.parameters(),
lr=c.lr_disc,
weight_decay=0)
# schedulers
scheduler_gen = None
scheduler_disc = None
if 'lr_scheduler_gen' in c:
scheduler_gen = getattr(torch.optim.lr_scheduler, c.lr_scheduler_gen)
scheduler_gen = scheduler_gen(optimizer_gen, **c.lr_scheduler_gen_params)
if 'lr_scheduler_disc' in c:
scheduler_disc = getattr(torch.optim.lr_scheduler, c.lr_scheduler_disc)
scheduler_disc = scheduler_disc(optimizer_disc, **c.lr_scheduler_disc_params)
# setup criterion
criterion_gen = GeneratorLoss(c)
criterion_disc = DiscriminatorLoss(c)
if args.restore_path:
checkpoint = torch.load(args.restore_path, map_location='cpu')
try:
print(" > Restoring Generator Model...")
model_gen.load_state_dict(checkpoint['model'])
print(" > Restoring Generator Optimizer...")
optimizer_gen.load_state_dict(checkpoint['optimizer'])
print(" > Restoring Discriminator Model...")
model_disc.load_state_dict(checkpoint['model_disc'])
print(" > Restoring Discriminator Optimizer...")
optimizer_disc.load_state_dict(checkpoint['optimizer_disc'])
if 'scheduler' in checkpoint:
print(" > Restoring Generator LR Scheduler...")
scheduler_gen.load_state_dict(checkpoint['scheduler'])
# NOTE: Not sure if necessary
scheduler_gen.optimizer = optimizer_gen
if 'scheduler_disc' in checkpoint:
print(" > Restoring Discriminator LR Scheduler...")
scheduler_disc.load_state_dict(checkpoint['scheduler_disc'])
scheduler_disc.optimizer = optimizer_disc
except RuntimeError:
# retore only matching layers.
print(" > Partial model initialization...")
model_dict = model_gen.state_dict()
model_dict = set_init_dict(model_dict, checkpoint['model'], c)
model_gen.load_state_dict(model_dict)
model_dict = model_disc.state_dict()
model_dict = set_init_dict(model_dict, checkpoint['model_disc'], c)
model_disc.load_state_dict(model_dict)
del model_dict
# reset lr if not countinuining training.
for group in optimizer_gen.param_groups:
group['lr'] = c.lr_gen
for group in optimizer_disc.param_groups:
group['lr'] = c.lr_disc
print(" > Model restored from step %d" % checkpoint['step'],
flush=True)
args.restore_step = checkpoint['step']
else:
args.restore_step = 0
if use_cuda:
model_gen.cuda()
criterion_gen.cuda()
model_disc.cuda()
criterion_disc.cuda()
# DISTRUBUTED
# if num_gpus > 1:
# model = apply_gradient_allreduce(model)
num_params = count_parameters(model_gen)
print(" > Generator has {} parameters".format(num_params), flush=True)
num_params = count_parameters(model_disc)
print(" > Discriminator has {} parameters".format(num_params), flush=True)
if 'best_loss' not in locals():
best_loss = float('inf')
global_step = args.restore_step
for epoch in range(0, c.epochs):
c_logger.print_epoch_start(epoch, c.epochs)
_, global_step = train(model_gen, criterion_gen, optimizer_gen,
model_disc, criterion_disc, optimizer_disc,
scheduler_gen, scheduler_disc, ap, global_step,
epoch)
eval_avg_loss_dict = evaluate(model_gen, criterion_gen, model_disc, criterion_disc, ap,
global_step, epoch)
c_logger.print_epoch_end(epoch, eval_avg_loss_dict)
target_loss = eval_avg_loss_dict[c.target_loss]
best_loss = save_best_model(target_loss,
best_loss,
model_gen,
optimizer_gen,
scheduler_gen,
model_disc,
optimizer_disc,
scheduler_disc,
global_step,
epoch,
OUT_PATH,
model_losses=eval_avg_loss_dict)
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument(
'--continue_path',
type=str,
help=
'Training output folder to continue training. Use to continue a training. If it is used, "config_path" is ignored.',
default='',
required='--config_path' not in sys.argv)
parser.add_argument(
'--restore_path',
type=str,
help='Model file to be restored. Use to finetune a model.',
default='')
parser.add_argument('--config_path',
type=str,
help='Path to config file for training.',
required='--continue_path' not in sys.argv)
parser.add_argument('--debug',
type=bool,
default=False,
help='Do not verify commit integrity to run training.')
# DISTRUBUTED
parser.add_argument(
'--rank',
type=int,
default=0,
help='DISTRIBUTED: process rank for distributed training.')
parser.add_argument('--group_id',
type=str,
default="",
help='DISTRIBUTED: process group id.')
args = parser.parse_args()
if args.continue_path != '':
args.output_path = args.continue_path
args.config_path = os.path.join(args.continue_path, 'config.json')
list_of_files = glob.glob(
args.continue_path +
"/*.pth.tar") # * means all if need specific format then *.csv
latest_model_file = max(list_of_files, key=os.path.getctime)
args.restore_path = latest_model_file
print(f" > Training continues for {args.restore_path}")
# setup output paths and read configs
c = load_config(args.config_path)
check_config(c)
_ = os.path.dirname(os.path.realpath(__file__))
OUT_PATH = args.continue_path
if args.continue_path == '':
OUT_PATH = create_experiment_folder(c.output_path, c.run_name,
args.debug)
AUDIO_PATH = os.path.join(OUT_PATH, 'test_audios')
c_logger = ConsoleLogger()
if args.rank == 0:
os.makedirs(AUDIO_PATH, exist_ok=True)
new_fields = {}
if args.restore_path:
new_fields["restore_path"] = args.restore_path
new_fields["github_branch"] = get_git_branch()
copy_config_file(args.config_path,
os.path.join(OUT_PATH, 'config.json'), new_fields)
os.chmod(AUDIO_PATH, 0o775)
os.chmod(OUT_PATH, 0o775)
LOG_DIR = OUT_PATH
tb_logger = TensorboardLogger(LOG_DIR, model_name='VOCODER')
# write model desc to tensorboard
tb_logger.tb_add_text('model-description', c['run_description'], 0)
try:
main(args)
except KeyboardInterrupt:
remove_experiment_folder(OUT_PATH)
try:
sys.exit(0)
except SystemExit:
os._exit(0) # pylint: disable=protected-access
except Exception: # pylint: disable=broad-except
remove_experiment_folder(OUT_PATH)
traceback.print_exc()
sys.exit(1)

0
vocoder/utils/__init__.py Normal file
View File

97
vocoder/utils/console_logger.py Normal file
View File

@ -0,0 +1,97 @@
import datetime
from TTS.utils.io import AttrDict
tcolors = AttrDict({
'OKBLUE': '\033[94m',
'HEADER': '\033[95m',
'OKGREEN': '\033[92m',
'WARNING': '\033[93m',
'FAIL': '\033[91m',
'ENDC': '\033[0m',
'BOLD': '\033[1m',
'UNDERLINE': '\033[4m'
})
class ConsoleLogger():
# TODO: merge this with TTS ConsoleLogger
def __init__(self):
# use these to compare values between iterations
self.old_train_loss_dict = None
self.old_epoch_loss_dict = None
self.old_eval_loss_dict = None
# pylint: disable=no-self-use
def get_time(self):
now = datetime.datetime.now()
return now.strftime("%Y-%m-%d %H:%M:%S")
def print_epoch_start(self, epoch, max_epoch):
print("\n{}{} > EPOCH: {}/{}{}".format(tcolors.UNDERLINE, tcolors.BOLD,
epoch, max_epoch, tcolors.ENDC),
flush=True)
def print_train_start(self):
print(f"\n{tcolors.BOLD} > TRAINING ({self.get_time()}) {tcolors.ENDC}")
def print_train_step(self, batch_steps, step, global_step,
step_time, loader_time, lrG, lrD,
loss_dict, avg_loss_dict):
indent = " | > "
print()
log_text = "{} --> STEP: {}/{} -- GLOBAL_STEP: {}{}\n".format(
tcolors.BOLD, step, batch_steps, global_step, tcolors.ENDC)
for key, value in loss_dict.items():
# print the avg value if given
if f'avg_{key}' in avg_loss_dict.keys():
log_text += "{}{}: {:.5f} ({:.5f})\n".format(indent, key, value, avg_loss_dict[f'avg_{key}'])
else:
log_text += "{}{}: {:.5f} \n".format(indent, key, value)
log_text += f"{indent}step_time: {step_time:.2f}\n{indent}loader_time: {loader_time:.2f}\n{indent}lrG: {lrG}\n{indent}lrD: {lrD}"
print(log_text, flush=True)
# pylint: disable=unused-argument
def print_train_epoch_end(self, global_step, epoch, epoch_time,
print_dict):
indent = " | > "
log_text = f"\n{tcolors.BOLD} --> TRAIN PERFORMACE -- EPOCH TIME: {epoch} sec -- GLOBAL_STEP: {global_step}{tcolors.ENDC}\n"
for key, value in print_dict.items():
log_text += "{}{}: {:.5f}\n".format(indent, key, value)
print(log_text, flush=True)
def print_eval_start(self):
print(f"{tcolors.BOLD} > EVALUATION {tcolors.ENDC}\n")
def print_eval_step(self, step, loss_dict, avg_loss_dict):
indent = " | > "
print()
log_text = f"{tcolors.BOLD} --> STEP: {step}{tcolors.ENDC}\n"
for key, value in loss_dict.items():
# print the avg value if given
if f'avg_{key}' in avg_loss_dict.keys():
log_text += "{}{}: {:.5f} ({:.5f})\n".format(indent, key, value, avg_loss_dict[f'avg_{key}'])
else:
log_text += "{}{}: {:.5f} \n".format(indent, key, value)
print(log_text, flush=True)
def print_epoch_end(self, epoch, avg_loss_dict):
indent = " | > "
log_text = " {}--> EVAL PERFORMANCE{}\n".format(
tcolors.BOLD, tcolors.ENDC)
for key, value in avg_loss_dict.items():
# print the avg value if given
color = ''
sign = '+'
diff = 0
if self.old_eval_loss_dict is not None and key in self.old_eval_loss_dict:
diff = value - self.old_eval_loss_dict[key]
if diff < 0:
color = tcolors.OKGREEN
sign = ''
elif diff > 0:
color = tcolors.FAIL
sign = '+'
log_text += "{}{}:{} {:.5f} {}({}{:.5f})\n".format(indent, key, color, value, tcolors.ENDC, sign, diff)
self.old_eval_loss_dict = avg_loss_dict
print(log_text, flush=True)

102
vocoder/utils/generic_utils.py Normal file
View File

@ -0,0 +1,102 @@
import re
import importlib
import numpy as np
from matplotlib import pyplot as plt
from TTS.utils.visual import plot_spectrogram
def plot_results(y_hat, y, ap, global_step, name_prefix):
""" Plot vocoder model results """
# select an instance from batch
y_hat = y_hat[0].squeeze(0).detach().cpu().numpy()
y = y[0].squeeze(0).detach().cpu().numpy()
spec_fake = ap.melspectrogram(y_hat).T
spec_real = ap.melspectrogram(y).T
spec_diff = np.abs(spec_fake - spec_real)
# plot figure and save it
fig_wave = plt.figure()
plt.subplot(2, 1, 1)
plt.plot(y)
plt.title("groundtruth speech")
plt.subplot(2, 1, 2)
plt.plot(y_hat)
plt.title(f"generated speech @ {global_step} steps")
plt.tight_layout()
plt.close()
figures = {
name_prefix + "spectrogram/fake": plot_spectrogram(spec_fake),
name_prefix + "spectrogram/real": plot_spectrogram(spec_real),
name_prefix + "spectrogram/diff": plot_spectrogram(spec_diff),
name_prefix + "speech_comparison": fig_wave,
}
return figures
def to_camel(text):
text = text.capitalize()
return re.sub(r'(?!^)_([a-zA-Z])', lambda m: m.group(1).upper(), text)
def setup_generator(c):
print(" > Generator Model: {}".format(c.generator_model))
MyModel = importlib.import_module('TTS.vocoder.models.' +
c.generator_model.lower())
MyModel = getattr(MyModel, to_camel(c.generator_model))
if c.generator_model in 'melgan_generator':
model = MyModel(
in_channels=c.audio['num_mels'],
out_channels=1,
proj_kernel=7,
base_channels=512,
upsample_factors=c.generator_model_params['upsample_factors'],
res_kernel=3,
num_res_blocks=c.generator_model_params['num_res_blocks'])
if c.generator_model in 'melgan_fb_generator':
pass
if c.generator_model in 'multiband_melgan_generator':
model = MyModel(
in_channels=c.audio['num_mels'],
out_channels=4,
proj_kernel=7,
base_channels=384,
upsample_factors=c.generator_model_params['upsample_factors'],
res_kernel=3,
num_res_blocks=c.generator_model_params['num_res_blocks'])
return model
def setup_discriminator(c):
print(" > Discriminator Model: {}".format(c.discriminator_model))
MyModel = importlib.import_module('TTS.vocoder.models.' +
c.discriminator_model.lower())
MyModel = getattr(MyModel, to_camel(c.discriminator_model))
if c.discriminator_model in 'random_window_discriminator':
model = MyModel(
cond_channels=c.audio['num_mels'],
hop_length=c.audio['hop_length'],
uncond_disc_donwsample_factors=c.
discriminator_model_params['uncond_disc_donwsample_factors'],
cond_disc_downsample_factors=c.
discriminator_model_params['cond_disc_downsample_factors'],
cond_disc_out_channels=c.
discriminator_model_params['cond_disc_out_channels'],
window_sizes=c.discriminator_model_params['window_sizes'])
if c.discriminator_model in 'melgan_multiscale_discriminator':
model = MyModel(
in_channels=1,
out_channels=1,
kernel_sizes=(5, 3),
base_channels=c.discriminator_model_params['base_channels'],
max_channels=c.discriminator_model_params['max_channels'],
downsample_factors=c.
discriminator_model_params['downsample_factors'])
return model
def check_config(c):
pass

63
vocoder/utils/io.py Normal file
View File

@ -0,0 +1,63 @@
import os
import torch
import datetime
def save_model(model, optimizer, scheduler, model_disc, optimizer_disc,
scheduler_disc, current_step, epoch, output_path, **kwargs):
model_state = model.state_dict()
model_disc_state = model_disc.state_dict()\
if model_disc is not None else None
optimizer_state = optimizer.state_dict()\
if optimizer is not None else None
optimizer_disc_state = optimizer_disc.state_dict()\
if optimizer_disc is not None else None
scheduler_state = scheduler.state_dict()\
if scheduler is not None else None
scheduler_disc_state = scheduler_disc.state_dict()\
if scheduler_disc is not None else None
state = {
'model': model_state,
'optimizer': optimizer_state,
'scheduler': scheduler_state,
'model_disc': model_disc_state,
'optimizer_disc': optimizer_disc_state,
'scheduler_disc': scheduler_disc_state,
'step': current_step,
'epoch': epoch,
'date': datetime.date.today().strftime("%B %d, %Y"),
}
state.update(kwargs)
torch.save(state, output_path)
def save_checkpoint(model, optimizer, scheduler, model_disc, optimizer_disc,
scheduler_disc, current_step, epoch, output_folder,
**kwargs):
file_name = 'checkpoint_{}.pth.tar'.format(current_step)
checkpoint_path = os.path.join(output_folder, file_name)
print(" > CHECKPOINT : {}".format(checkpoint_path))
save_model(model, optimizer, scheduler, model_disc, optimizer_disc,
scheduler_disc, current_step, epoch, checkpoint_path, **kwargs)
def save_best_model(target_loss, best_loss, model, optimizer, scheduler,
model_disc, optimizer_disc, scheduler_disc, current_step,
epoch, output_folder, **kwargs):
if target_loss < best_loss:
file_name = 'best_model.pth.tar'
checkpoint_path = os.path.join(output_folder, file_name)
print(" > BEST MODEL : {}".format(checkpoint_path))
save_model(model,
optimizer,
scheduler,
model_disc,
optimizer_disc,
scheduler_disc,
current_step,
epoch,
checkpoint_path,
model_loss=target_loss,
**kwargs)
best_loss = target_loss
return best_loss