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
4,800 Commits 80 Branches 98 Tags 314 MiB
Commit Graph

254 Commits

Author SHA1 Message Date
Eren Gölge f7587fc134
Fix SSIM loss correction 2022-07-13 10:47:12 +02:00
Eren Gölge bc1f93c299
Fix device allocation 2022-07-12 19:05:25 +02:00
Eren Gölge 49bac724c0
Implement VitsAudioConfig (#1556) 
* Implement VitsAudioConfig

* Update VITS LJSpeech recipe

* Update VITS VCTK recipe

* Make style

* Add missing decorator

* Add missing param

* Make style

* Update recipes

* Fix test

* Bug fix

* Exclude tests folder

* Make linter

* Make style
 2022-07-12 18:49:58 +02:00
Eren G??lge 48a4f3647f Make lint 2022-07-12 14:58:26 +02:00
Eren G??lge 2cf89b88c9 Make style 2022-07-12 14:12:57 +02:00
Eren G??lge a6f73a18cb Fix BCELoss adressing #1192 2022-07-12 14:11:34 +02:00
Eren G??lge c17ff17a18 Fix SSIM loss 2022-07-12 12:35:24 +02:00
a-froghyar 8be21ec387
Capacitron (#977) 
* new CI config

* initial Capacitron implementation

* delete old unused file

* fix empty formatting changes

* update losses and training script

* fix previous commit

* fix commit

* Add Capacitron test and first round of test fixes

* revert formatter change

* add changes to the synthesizer

* add stepwise gradual lr scheduler and changes to the recipe

* add inference script for dev use

* feat: add posterior inference arguments to synth methods
- added reference wav and text args for posterior inference
- some formatting

* fix: add espeak flag to base_tts and dataset APIs
- use_espeak_phonemes flag was not implemented in those APIs
- espeak is now able to be utilised for phoneme generation
- necessary phonemizer for the Capacitron model

* chore: update training script and style
- training script includes the espeak flag and other hyperparams
- made style

* chore: fix linting

* feat: add Tacotron 2 support

* leftover from dev

* chore:rename parser args

* feat: extract optimizers
- created a separate optimizer class to merge the two optimizers

* chore: revert arbitrary trainer changes

* fmt: revert formatting bug

* formatting again

* formatting fixed

* fix: log func

* fix: update optimizer
- Implemented load_state_dict for continuing training

* fix: clean optimizer init for standard models

* improvement: purge espeak flags and add training scripts

* Delete capacitronT2.py

delete old training script, new one is pushed

* feat: capacitron trainer methods
- extracted capacitron specific training  operations from the trainer into custom
methods in taco1 and taco2 models

* chore: renaming and merging capacitron and gst style args

* fix: bug fixes from the previous commit

* fix: implement state_dict method on CapacitronOptimizer

* fix: call method

* fix: inference naming

* Delete train_capacitron.py

* fix: synthesize

* feat: update tests

* chore: fix style

* Delete capacitron_inference.py

* fix: fix train tts t2 capacitron tests

* fix: double forward in T2 train step

* fix: double forward in T1 train step

* fix: run make style

* fix: remove unused import

* fix: test for T1 capacitron

* fix: make lint

* feat: add blizzard2013 recipes

* make style

* fix: update recipes

* chore: make style

* Plot test sentences in Tacotron

* chore: make style and fix import

* fix: call forward first before problematic floordiv op

* fix: update recipes

* feat: add min_audio_len to recipes

* aux_input["style_mel"]

* chore: make style

* Make capacitron T2 recipe more stable

* Remove T1 capacitron Ljspeech

* feat: implement new grad clipping routine and update configs

* make style

* Add pretrained checkpoints

* Add default vocoder

* Change trainer package

* Fix grad clip issue for tacotron

* Fix scheduler issue with tacotron

Co-authored-by: Eren Gölge <egolge@coqui.ai>
Co-authored-by: WeberJulian <julian.weber@hotmail.fr>
Co-authored-by: Eren Gölge <erogol@hotmail.com>
 2022-05-20 16:17:11 +02:00
code-review-doctor fa887ef5f9
Fix issue probably-meant-fstring found at https://codereview.doctor (#1532) 2022-05-07 13:33:40 +02:00
Edresson Casanova 8d228ab22a
Trick to Upsampling to High sampling rates using VITS model (#1456) 
* Add upsample VITS support

* Fix the bug in inference

* Fix lint checks

* Add RMS based norm in save_wav method

* Style fix

* Add the period for VITS multi-period discriminator in model_args

* Bug fix in speaker encoder load in inference time

* Add unit tests

* Remove useless detach_z_vocoder parameter

* Add docs for VITS upsampling

* Fix the docs

* Rename TTS_part_sample_rate to encoder_sample_rate

* Add upsampling_init and upsampling_z methods

* Add asserts for encoder_sample_rate part

* Move upsampling tests to test_vits.py
 2022-04-26 11:47:46 +02:00
Eren Gölge 424d04e4f6 Make stlye 2022-02-25 11:31:56 +01:00
Eren Gölge 52a7896668 Update VITS loss 2022-02-25 11:30:24 +01:00
Eren Gölge 1a43e05460 Fix VITS loss bug 
Fake and real features were given in the wrong args order to
the loss function
 2022-02-25 11:26:59 +01:00
Eren Gölge 1f0c8179da Make style 2022-02-25 11:26:59 +01:00
Eren Gölge 34c4be5e49 Update forwardtts 2022-02-25 11:26:59 +01:00
Eren Gölge 146fbfd7c9 Extend unittests 2022-02-25 11:25:00 +01:00
Eren Gölge 127118c637
Update TTS.tts formatters (#1228) 
* Return Dict from tts formatters

* Make style
 2022-02-11 23:03:43 +01:00
Edresson Casanova 0860d73cf8
Remove Tensorflow requeriment (#1225) 
* Remove TF modules

* Remove TF unit tests

* Remove TF vocoder modules

* Remove TF convert scripts

* Remove TF requirement

* Remove the Docs TF instructions

* Remove TF inference support
 2022-02-10 16:14:54 +01:00
Eren Gölge 704dddcffa Make style 2021-12-20 11:54:10 +00:00
Edresson 12968532fe Add the language embedding dim in the duration predictor class 2021-12-20 11:54:10 +00:00
Edresson 8c22d5ac49 Turn more clear the VITS loss function 2021-12-20 11:54:10 +00:00
Edresson 6fc3b9e679 Remove the unusable fine-tuning model 2021-12-20 11:54:10 +00:00
WeberJulian 1472b6df49 make style 2021-12-20 11:54:10 +00:00
Edresson eeb8ac07d9 Add voice conversion fine tuning mode 2021-12-20 11:54:10 +00:00
Edresson 690b37d0ab Add support to use the speaker encoder as loss function in VITS model 2021-12-20 11:54:09 +00:00
Edresson c53693c155 Implement vocoder Fine Tuning like SC-GlowTTS paper 2021-12-20 11:54:09 +00:00
Edresson dcb2374bc9 Add multilingual training support to the VITS model 2021-12-20 11:54:09 +00:00
Eren Gölge b6b14a76af Fix VITS stochastic duration predictor 2021-11-08 09:20:11 +01:00
Eren Gölge 0e768dd4c5 Update comments 2021-10-20 18:21:26 +00:00
Eren Gölge fd95926009 Update GlowTTS 2021-09-30 14:47:56 +00:00
Eren Gölge 2766dd1d6e
Fix #813 - GlowTTS training (#814) 
* Fix #813

* Update glow_tts recipe

* Fix glow-tts test

* Linter fix

* Run data dep init only in training
 2021-09-17 20:06:55 +02:00
Eren Gölge 26f76fce22 Remove SpeedySpeech from .models.json 2021-09-10 17:47:27 +00:00
Eren Gölge d6e29ef98a Style update 2021-09-10 08:30:33 +00:00
Eren Gölge 570d5971be Implement `ForwardTTSLoss` 2021-09-10 08:29:12 +00:00
Eren Gölge bfc6ceac29 Move MAS to `TTS.tts.utils.helpers` 2021-09-09 10:57:19 +00:00
Eren Gölge 4761853c5c Fix imports 2021-09-08 13:34:40 +00:00
Eren Gölge 2b59da802c Fix loader setup in `base_tts` 2021-09-06 15:16:58 +00:00
Eren Gölge 29248536c9 Update `PositionalEncoding` 2021-09-06 15:16:58 +00:00
Eren Gölge 4672889549 Update `generic.FFTransformer` 2021-09-06 15:16:58 +00:00
Eren Gölge 2bf9e83c49 FastPitch refactor and commenting 2021-09-06 15:16:58 +00:00
Eren Gölge 59b24e66cf Add `AlignerNetwork` 2021-09-06 15:16:58 +00:00
Eren Gölge debf772ec5 Implement binary alignment loss 2021-09-06 15:16:58 +00:00
Eren Gölge e429afbce4 Enable aligner for FastPitch 2021-09-06 15:16:58 +00:00
Eren Gölge fac9dbe661 Update FastPitchLoss 2021-09-06 15:16:58 +00:00
Eren Gölge b81560607b Update docstrings 2021-09-06 15:16:58 +00:00
Eren Gölge 8fffd4e813 Don't print computed phonemes 
It causes noise in logs
 2021-09-06 15:16:58 +00:00
Eren Gölge db32162eae Fix `FastPitchLoss` 2021-09-06 15:16:58 +00:00
Eren Gölge c8d999b010 Add FastPitchLoss 2021-09-06 15:16:58 +00:00
Eren Gölge 18da8f5dbd Update pylint 2.10.2 and fix lint issues 2021-08-30 08:10:35 +00:00
Eren Gölge 2620f62ea8 Move duration_loss inside VitsGeneratorLoss 2021-08-27 07:07:07 +00:00
Eren Gölge 49e1181ea4 Fixes for the vits model 2021-08-26 17:15:09 +00:00
Eren Gölge 3ab8cef99e Fix VITS model SPD 2021-08-18 14:55:46 +00:00
Eren Gölge c312acac7d Implement VITS model 🚀 
VITS model implementation built on Glow TTS and HiFiGAN
layers.
 2021-08-09 18:02:36 +00:00
Eren Gölge e4648ffef1 Fix multi-speaker init of Tacotron models & tests 2021-08-09 18:02:36 +00:00
Eren Gölge fc0c4600bd Fix stopnet training 2021-07-24 11:39:54 +02:00
Eren Gölge 2e1a428b83 Update glowtts docstrings and docs 2021-06-30 14:30:55 +02:00
Eren Gölge ae6405bb76 Docstrings for `Trainer` 2021-06-28 17:03:47 +02:00
Eren Gölge d42d1c02ea Use `torch.linalg.qr` for pytorch > `v1.9.0` 2021-06-28 17:03:47 +02:00
Eren Gölge a5d5bc9063 Print `max_decoder_steps` when model reaches the limit 2021-06-28 17:03:47 +02:00
Eren Gölge 269e5a734e add max_decoder_steps argument to tacotron models 2021-06-28 17:03:19 +02:00
Eren Gölge db6a97d1a2 rename external speaker embedding arguments as `d_vectors` 2021-06-28 17:03:19 +02:00
Eren Gölge b500338faa make style 2021-06-28 17:03:19 +02:00
Eren Gölge 9203b863d9 update align_tts_loss for trainer 2021-06-28 17:03:19 +02:00
Eren Gölge 9134c7dfb6 update `sequence_mask` import globally 2021-06-28 17:03:19 +02:00
Eren Gölge ca302db7b0 add sequence_mask to `utils.data` 2021-06-28 17:03:19 +02:00
Eren Gölge 844abb3b1d `setup_loss()` in `layer/__init__.py` 2021-06-28 17:03:19 +02:00
Adam Froghyar 7ddc885f37 deleted a line the broke GravesAttention 2021-05-10 15:42:59 +02:00
Eren Gölge 8cb27267a4 formatting 2021-05-03 14:26:35 +02:00
Eren Gölge 9cc17be53a formatting and a small bug fix in Tacotron model 2021-04-15 16:36:51 +02:00
Eren Gölge 3de5a89154 optionally enable prenet dropout at inference time for tacotron models 2021-04-13 13:24:56 +02:00
Eren Gölge 87ee6ceb57 style update #3 2021-04-09 01:17:15 +02:00
Eren Gölge 18d9ec8036 format with black 2021-04-09 00:54:59 +02:00
Eren Gölge e5b9607bc3 isort all imports 2021-04-09 00:45:20 +02:00
Eren Gölge 0e79fa86ad format with black and pylint 2.7.3 2021-04-09 00:38:08 +02:00
Eren Gölge 44b4cb5ba5 DCA comment 2021-04-06 16:24:50 +02:00
Eren Gölge a3a840fd78 linter fixes 2021-03-30 14:39:16 +02:00
Eren Gölge 6b2e13bf62 compute normalized logp using torch primitives 2021-03-30 14:39:16 +02:00
Eren Gölge 7a382a5c2b stowed aligntts commit and small refactoring with feed_forward layers 2021-03-30 14:39:16 +02:00
Eren Gölge d542a50818 fix losses for alignTTS 2021-03-30 14:39:16 +02:00
Eren Gölge 18cc7b95ec update l1 and huber to mse loss 2021-03-30 14:39:16 +02:00
Eren Gölge 896d33ed49 update losses to hande alingtts phases 2021-03-30 14:39:16 +02:00
Eren Gölge c2d29e5cd4 FFTransformer encoder for aligntts 2021-03-30 14:39:16 +02:00
Eren Gölge 460a2d3e26 FFTransformer Decoder for AlignTTS 2021-03-30 14:39:16 +02:00
Eren Gölge aa29f5b199 aligntts loss 2021-03-30 14:39:16 +02:00
Eren Gölge a831468cab align tts MDN layer 2021-03-30 14:39:16 +02:00
Eren Gölge 4396f8e2da continue refactoring 2021-03-30 14:39:16 +02:00
Eren Gölge 2b3e12ea49 correct imports after refactoring, add AlignTTS (old SSMAS) and some formatting 2021-03-30 14:39:16 +02:00
Eren Gölge d9c405f0c3 create feedforward folder for SS layers 2021-03-30 14:39:16 +02:00
Eren Gölge a8cf1ae6b4 fix wavenet running with no input mask 2021-03-30 14:39:16 +02:00
Eren Gölge 32e8b56c45 linter fix 2021-03-18 13:33:23 +01:00
Eren Gölge 65533f33e9 fix #374 2021-03-18 13:33:00 +01:00
Eren Gölge 9a48ba3821 a ton of linter updates 2021-03-08 05:06:54 +01:00
Eren Gölge 08581deb61 linter updates 2021-03-08 02:53:02 +01:00
Eren Gölge b464cab9b8 setup.py update and pylint fixes 2021-01-26 02:57:50 +01:00
Eren Gölge 660d61aeeb maximum_path_numpy and CYTHON adabtable import 2021-01-26 02:57:07 +01:00
root 5c87753e88 glow-tts fix for saving inverse weight 2021-01-20 02:09:42 +00:00
erogol bbc8d665a1 move attention layers to a sperate file 2021-01-11 17:27:30 +01:00
erogol 79c841ccd3 mass refactoring and update 2021-01-11 17:26:58 +01:00
erogol 1d961d6f8a cladd renaming 2021-01-11 17:26:11 +01:00
erogol c0a2aa68d3 formatting 2021-01-11 17:25:39 +01:00
erogol b206162d11 more docstrings 2021-01-11 17:25:04 +01:00
erogol 6e9043c5d2 rename convbnblocks and handle none mask 2021-01-11 17:22:34 +01:00
erogol 921fa5db92 remove attentions from common layers 2021-01-11 15:06:42 +01:00
erogol cc2b1e043d docstrings for common layers 2021-01-11 15:06:12 +01:00
erogol a6f40fef2e stage missing files 2021-01-08 16:02:56 +01:00
erogol d382d759b3 small fixes and test fixes 2021-01-08 15:48:40 +01:00
erogol a6259041d3 docstring for speedyspeech 2021-01-07 14:35:22 +01:00
erogol de2a542f83 glow-tts bug fix 2021-01-07 13:40:32 +01:00
erogol 5a45af48f1 fix 2021-01-06 13:19:40 +01:00
erogol e7fad928e7 doc strings for the all glow-tts layers 2021-01-06 13:19:40 +01:00
erogol d3b7284be4 glow-tts comments and refactoring 2021-01-06 13:19:40 +01:00
erogol 7586fbc4de SS refactoring 2021-01-06 13:19:40 +01:00
erogol e82d31b6ac glow ttss refactoring 2021-01-06 13:19:40 +01:00
erogol 29f4329d7f update glow-tts layers and add some comments 2021-01-06 13:19:40 +01:00
erogol eb555855e4 small fixes 2021-01-06 13:19:40 +01:00
erogol 5901a00576 argument rename 2021-01-06 13:19:40 +01:00
erogol 4ef083f0f1 select decoder type for SS 2021-01-06 13:19:40 +01:00
erogol 3fa408a5ea change order BN + ReLU to ReLU + BN for SS 2021-01-06 13:19:40 +01:00
erogol ac5c9217d1 positional encoding masking for SS 2021-01-06 13:19:40 +01:00
erogol fede46e96e pylint and test fixes 2021-01-06 13:19:40 +01:00
erogol cf869e8922 add SS files 2021-01-06 13:19:40 +01:00
erogol dc4a16d62e speedy speehc losses 2021-01-06 13:19:40 +01:00
erogol d62cac7252 fix glow-tts prenet bug fix 2021-01-06 13:19:40 +01:00
erogol fa6907fa0e update glow-tts parameters and fix rel-attn-win size 2021-01-06 13:19:40 +01:00
erogol 7b20d8cbd3 implement residual BN convolution and add it as an alternative encoder for glow-tts. also generic layers to layers/generic 2021-01-06 13:19:40 +01:00
erogol 070146e143 add monotonic dynamic convolution attention 2021-01-06 13:18:41 +01:00
erogol f6c96b0ac2 Merge branch 'dev' 2020-11-25 15:29:06 +01:00
erogol d94782a076 reset the way ga_loss is stored in return_dict 2020-11-02 13:18:56 +01:00
erogol a108d0ee81 check nan loss in glow-tts loss 2020-11-02 13:12:19 +01:00
erogol b8ac9aba9d check against NaN loss in tacotron_loss 2020-11-02 12:44:41 +01:00
erogol 183fe56d95 Merge branch 'ssim_loss' into dev 2020-10-29 23:49:09 +01:00
erogol 946a0c0fb9 bug fixes for single speaker glow-tts, enable torch based amp. Make amp optional for wavegrad. Bug fixes for synthesis setup for glow-tts 2020-10-29 15:45:50 +01:00
erogol fdaed45f58 optional loss masking for stoptoken predictor 2020-10-28 18:40:54 +01:00
erogol e49cc3bbcd bug fix 2020-10-28 18:34:34 +01:00
erogol 9cef923d99 ssim loss for tacotron models 2020-10-28 15:24:18 +01:00
erogol a6f564c8c8 pylint fixes 2020-10-27 12:35:10 +01:00
erogol 8de7c13708 fix no loss masking loss computation 2020-10-27 12:17:38 +01:00
Alexander Korolev 47d74ced1c
Update losses.py 
Seems like in the latest dev merge, this change was reverted. Any specific reason for this?
Without it the problem as stated here https://github.com/mozilla/TTS/issues/473 occurs.
 2020-10-23 14:15:01 +02:00
erogol c2c4126a18 remove merge conflicts 2020-10-08 01:35:27 +02:00
erogol 6f0654f9a8 differential spectral loss 2020-10-08 01:30:42 +02:00
erogol 4e93f90108 bug fix 2020-10-08 01:29:30 +02:00
erogol bb9b70ee27 differential spectral loss and loss weight settings 2020-10-08 01:29:30 +02:00
Edresson 99d5a0ac07 add Speaker Conditional GST support 2020-09-29 16:09:27 -03:00
erogol 6a70c63f24 correct glow-tts loss 2020-09-27 03:28:42 +02:00
erogol 665f7ca714 linter fix 2020-09-24 12:57:54 +02:00
erogol 10258724d1 linter fixes 2020-09-22 03:54:16 +02:00
erogol e0b9fa887f glow-tts modules added 2020-09-21 14:15:40 +02:00
erogol e4c6386603 change import for normalization layer 2020-09-21 13:09:52 +02:00
erogol c008003506 do not check sample rate as loading stats file for normalization to enable interpolation for different sample rate vocoder 2020-09-18 12:52:19 +02:00
erogol 3660c57f1e time seperable convolution encoder, huber loss for duration predictor 2020-09-17 03:10:58 +02:00