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
2,927 Commits 80 Branches 98 Tags 314 MiB
Commit Graph

763 Commits

Author SHA1 Message Date
Edresson 77d85c6cc5 add softmaxproto loss and bug fix in data loader 2021-05-10 17:08:38 -03:00
chmodsss 607d5cf377 [#480] Adding version variable 2021-05-10 19:46:34 +02:00
Adam Froghyar 7ddc885f37 deleted a line the broke GravesAttention 2021-05-10 15:42:59 +02:00
Edresson 78bad25f2b update voxceleb download link 2021-05-07 23:45:15 -03:00
Eren Gölge f7582107da
Merge pull request #453 from Edresson/dev 
Script for spectrogram extraction using teacher forcing and Glow-TTS inference with MAS.
 2021-05-06 17:53:28 +02:00
Edresson 501c8e0302 remove unused vars on extract tts spectrograms script 2021-05-04 19:04:13 -03:00
Eren Gölge 0325c58862
Merge pull request #468 from shaun95/patch-1 
Update losses.py
 2021-05-03 14:45:24 +02:00
Eren Gölge 8cb27267a4 formatting 2021-05-03 14:26:35 +02:00
Eren Gölge 87d674a038 bumpup librosa version to 0.8.0 2021-05-03 14:25:09 +02:00
shaun 7d0ec62bf1
Update losses.py 
The block of code for use_l1_spec_loss is repeated which doubles the amount of L1 loss when enabled.
The weight for L1 loss in hifigan_ljspeech configutation will likely need to be doubled to compensate (l1_spec_loss_weight)
 2021-05-02 14:14:24 +02:00
Edresson 3ecd556bbe add unit test for extract tts spectrograms script 2021-05-01 13:41:56 -03:00
Edresson 446b1da936 create inference function 2021-04-29 18:18:37 -03:00
Eren Gölge f02f0338c2 fix .models.json and add testing to check released models availability 2021-04-29 09:32:36 +02:00
Eren Gölge fd95e9b8a4 [ci skip] Add sam models 2021-04-28 21:57:31 +02:00
Agrin Hilmkil 351d0ed6ae Remove unnecessary fsspec usage 2021-04-28 11:21:08 +02:00
Agrin Hilmkil 167f86417e Move dev, tf, notebook dependencies to extras 2021-04-28 11:20:06 +02:00
Eren Gölge 1235e54738 test for synthesize.py 2021-04-27 14:17:38 +02:00
Eren Gölge 4719414f2e remove imports 2021-04-27 11:25:17 +02:00
Eren Gölge add97cddc1 move function and remove import 2021-04-27 11:22:56 +02:00
Eren Gölge 734e6a515c bug fix 2021-04-27 10:27:45 +02:00
Eren Gölge 6bdd81667e place holders for sc-glow and hifigan models 2021-04-26 19:53:12 +02:00
Eren Gölge 2f0716073e enable multi-speaker CoquiTTS models for synthesize.py 2021-04-26 19:36:53 +02:00
Eren Gölge b531fa699c remove conflicy noise 2021-04-26 15:27:52 +02:00
Eren Gölge f37b488876 Merge branch 'speaker-manager' of https://github.com/coqui-ai/TTS into speaker-manager 2021-04-26 15:25:25 +02:00
Eren Gölge b82daa5e86 style and linter fixes 2021-04-26 15:22:24 +02:00
Edresson 20e42a3381 add save audio option 2021-04-23 15:00:00 -03:00
Edresson 8228091f92 add script for extraction of tts spectrograms 2021-04-23 14:17:46 -03:00
Eren Gölge 4cf211348d styling and linting 2021-04-23 18:04:37 +02:00
Eren Gölge 7eb0c60d2e let synthesizer to pass speaker encoder file paths to speaker manager 2021-04-23 18:04:37 +02:00
Eren Gölge f69195739e let speaker manager compute mean x_vector from multiple wav files 2021-04-23 18:04:37 +02:00
Eren Gölge 179722e3a7 new arguments to synthesize.py for loading speaker encoder and speaker wavs 2021-04-23 18:04:37 +02:00
Eren Gölge dfa415a8b8 small refactor in server.py 2021-04-23 18:04:37 +02:00
Eren Gölge c80d21f311 load speaker_encoder_ap and compute x_vector directly from the input file in speaker manager 2021-04-23 18:04:37 +02:00
Eren Gölge ad047c8195 html formatting, enable multi-speaker model on the server with a dropdown menu to select the speaker 2021-04-23 18:04:37 +02:00
Eren Gölge f9f3d04d14 remove moved function 2021-04-23 18:04:37 +02:00
Eren Gölge 10c988ac8c update server.py 2021-04-23 18:04:37 +02:00
Eren Gölge 6d0f5e0459 use SpeakerManager in Synthesizer 2021-04-23 18:04:37 +02:00
Eren Gölge e97126314c add ```unique``` argument to make_symbols to fix the incompat. issue of the 
SC-Glow models
 2021-04-23 18:04:37 +02:00
Eren Gölge d08888e603 formating speakers.py 2021-04-23 18:04:37 +02:00
Eren Gölge df422223a3 initial SpeakerManager implementation 2021-04-23 18:04:37 +02:00
Eren Gölge 7a7aeb35f5 fix the glow-tts in setup_model 2021-04-23 18:04:37 +02:00
Eren Gölge d42748082a update argument name external_speaker_embedding_dim -> speaker_embedding_dim 
add inference_noise_scale argument to glow-tts
 2021-04-23 18:04:37 +02:00
Eren Gölge 2da81f5bb6 add load_chekpoint to speaker encoder 2021-04-23 18:04:37 +02:00
Eren Gölge 1229ccbf07 update argument name in server.py 2021-04-23 18:04:37 +02:00
Eren Gölge af2d36faeb update synthesize.py for multi-speaker setting 2021-04-23 18:04:37 +02:00
Eren Gölge 99dc07a7dd add ```unique``` param to keep scglow models compatible (they are duplicate symbols ins the character set) 2021-04-23 18:04:37 +02:00
Eren Gölge c955a12428 set the default layer size compatible with scglow 2021-04-23 18:04:37 +02:00
Eren Gölge 3ace2440fa fix a mistake from rebase 2021-04-23 18:04:37 +02:00
Eren Gölge aadb2106ec code styling 2021-04-23 18:04:37 +02:00
Eren Gölge af7baa3387 refactoring to allow defining the speaker file externally 2021-04-23 18:04:37 +02:00
kirianguiller 7dccbfdcd5 handle multi speaker and gst in Synthetizer class 2021-04-23 18:04:37 +02:00
Edresson d2b6326b8b change optimizer initialization for compatibility with Hifi-GAN official implementation 2021-04-23 07:54:39 -03:00
WeberJulian 4205284f92
Change name of the functions 2021-04-23 10:09:55 +02:00
WeberJulian a26498181b Change back the default value 2021-04-22 16:10:17 +02:00
Julian Weber 355e1f47ab fix dumb mistake 2021-04-22 15:50:29 +02:00
Julian Weber c125b71f36 fix windows support 2021-04-22 15:14:24 +02:00
Jörg Thalheim f5fd7f78d4 server: also listen to ipv6 
The [::] address will listen to both ipv4/ipv6 addresses.
 2021-04-22 12:38:55 +02:00
Eren Gölge ef37633cb3 [ci skip] use prenet_dropout by default with Tacotron models 2021-04-22 12:38:55 +02:00
Eren Gölge e1d960da9e use SpeakerManager in Synthesizer 2021-04-21 13:13:27 +02:00
Eren Gölge 04b6881b66 add ```unique``` argument to make_symbols to fix the incompat. issue of the 
SC-Glow models
 2021-04-21 13:12:35 +02:00
Eren Gölge 790946faec formating speakers.py 2021-04-21 13:12:11 +02:00
Eren Gölge ab313814de initial SpeakerManager implementation 2021-04-21 13:11:46 +02:00
Eren Gölge 09890c7421 fix the glow-tts in setup_model 2021-04-21 13:10:40 +02:00
Eren Gölge 8764d02eb2 update argument name external_speaker_embedding_dim -> speaker_embedding_dim 
add inference_noise_scale argument to glow-tts
 2021-04-21 13:09:44 +02:00
Eren Gölge 8b40720977 add load_chekpoint to speaker encoder 2021-04-21 13:09:04 +02:00
Eren Gölge 37cad38c27 update argument name in server.py 2021-04-21 13:08:45 +02:00
Eren Gölge 9bccee9da8 update synthesize.py for multi-speaker setting 2021-04-21 13:08:25 +02:00
Eren Gölge d2fa8add1f add ```unique``` param to keep scglow models compatible (they are duplicate symbols ins the character set) 2021-04-16 19:40:13 +02:00
Eren Gölge d9612a4351 set the default layer size compatible with scglow 2021-04-16 19:40:13 +02:00
Eren Gölge 1038fd420d fix a mistake from rebase 2021-04-16 19:39:47 +02:00
Eren Gölge 47e356cb48 code styling 2021-04-16 16:01:40 +02:00
Eren Gölge 25328aad00 refactoring to allow defining the speaker file externally 2021-04-16 15:59:57 +02:00
kirianguiller 48ae52a9a3 handle multi speaker and gst in Synthetizer class 2021-04-16 15:54:49 +02:00
Eren Gölge a53958ae3a fix urls for the new models 2021-04-15 17:05:00 +02:00
Eren Gölge 9cc17be53a formatting and a small bug fix in Tacotron model 2021-04-15 16:36:51 +02:00
Eren Gölge 1ad838bc83 add newly released models under .model.json 2021-04-15 16:06:10 +02:00
Eren Gölge 7cada1a949 remove noise 2021-04-15 15:30:45 +02:00
Eren Gölge d60a8d7211 show the real waveform on TB too for GAN vocoder training. 2021-04-15 15:30:06 +02:00
Eren Gölge 5fbe926429 change the default TTS model to TacotronDDC 2021-04-15 15:29:44 +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 28a2fed8a3 update hifigan in .model.json 2021-04-12 16:48:05 +02:00
Eren Gölge abaf36861a aligntts model .model.json placeholder 2021-04-12 16:43:52 +02:00
Eren Gölge 480e2f7888 docstring update and better handling make_symbols 2021-04-12 16:40:49 +02:00
Eren Gölge b735076bb4 linter fixes 2021-04-12 13:14:11 +02:00
Eren Gölge b11d1cb845 small fixes 2021-04-12 12:40:55 +02:00
Eren Gölge a7f6045644 Merge branch 'reformat' into hifigan-reformat 2021-04-12 12:00:17 +02:00
Eren Gölge f519012dea reformatting and styling 2021-04-12 11:47:39 +02:00
Eren Gölge 9011dddf77 tacotron DDC placeholder in models.json 2021-04-12 04:06:27 +02:00
Eren Gölge d295d5de97 remove torch.no_grad from TorchSTFT 2021-04-10 19:43:57 +02:00
Eren Gölge 5b70da2e3f restore schedulers only if training is continuing a previous training 
inherit nn.Module for TorchSTFT
 2021-04-09 19:31:28 +02:00
Eren Gölge 2c71c6d8cd [ci skip]update gan vocoder configs to reflect the recent changes 2021-04-09 17:15:32 +02:00
Eren Gölge 2b529f60c8 update default hifigan config 2021-04-09 11:40:06 +02:00
Eren Gölge 105e0b4d62 vocoder gan training fixes 2021-04-09 11:38:04 +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 cd69da4868 linter fixes #2 2021-04-08 16:57:46 +02:00
Eren Gölge 4d3e1e9d9a linter fix 2021-04-08 14:57:46 +02:00
Eren Gölge 53f54898bc small fixes 2021-04-08 14:22:47 +02:00
Eren Gölge 006b1d3aaa bug fix 2021-04-08 13:17:45 +02:00
Eren Gölge 3f0993aebe remove junk 2021-04-08 12:17:02 +02:00
Eren Gölge 0ee0458309 remove redundant imports 2021-04-08 11:29:15 +02:00
Eren Gölge 773f1db6fa refactor HifiGAN discriminator 2021-04-08 11:28:30 +02:00
Eren Gölge 15f362d5b1 formatting 2021-04-08 11:28:30 +02:00
Eren Gölge aee24b0704 set different seed in gan_dataset when it is multi-workers 2021-04-08 11:28:30 +02:00
Eren Gölge 6ee211c137 remove stft params causing warning 2021-04-08 11:28:30 +02:00
Eren Gölge 4998ece8d8 allow configuration of optimziers from the config file 2021-04-08 11:28:30 +02:00
Eren Gölge 8daf407652 cache empty 2021-04-08 11:28:30 +02:00
Eren Gölge 3fb78c004a move scheduler updates to the end of the epoch 2021-04-08 11:28:30 +02:00
Eren Gölge 2a872c98aa don't call os.exit as it leaves the process resources standing 2021-04-08 11:27:40 +02:00
Eren Gölge 7cecd2fb2e add hifigan D 2021-04-08 11:27:40 +02:00
Eren Gölge 13dca6e6b6 revert some of Hifigan generator updates 2021-04-08 11:27:40 +02:00
Eren Gölge 02bc776c35 prevenet grad in TorchSTFT 2021-04-08 11:27:40 +02:00
Eren Gölge cf44624df8 more docstring 2021-04-08 11:27:40 +02:00
Eren Gölge d95b1458e8 Linter fixes and docstrings for HiFiGAN 2021-04-08 11:27:40 +02:00
Eren Gölge bd7a1c177b fix #419 2021-04-08 11:26:41 +02:00
Eren Gölge 7726dfca99 change the upper bound in sound normalization 2021-04-08 11:26:01 +02:00
Eren Gölge 57f6bd1afa make using different samples for G and D networks optional 2021-04-08 11:26:01 +02:00
Eren Gölge 67f8248492 placeholder for finetuned sam hifigan model 2021-04-08 11:25:29 +02:00
Eren Gölge 241e968df1 load_checkpoint for hifigan and no_grad for inference 2021-04-08 11:25:29 +02:00
Eren Gölge de3a04f104 some commeting for Generator loss and check if the argument is defines in the config file 2021-04-08 11:25:29 +02:00
Eren Gölge ff07c5f5e3 update TorchSTFT to enable melspec 2021-04-08 11:25:29 +02:00
Eren Gölge 4a5b1d4ac2 update hifigan config 2021-04-08 11:24:21 +02:00
Eren Gölge e0e3b12b26 pass all parameters explicity to _istft 2021-04-08 11:23:20 +02:00
Eren Gölge f0e76ee135 initial models.json entry for universal hifigan 2021-04-08 11:23:20 +02:00
Eren Gölge d57f416957 small fixes 2021-04-08 11:22:30 +02:00
Eren Gölge 8c9e1c9e58 hifigan implementation update 2021-04-08 11:21:43 +02:00
Eren Gölge a14d7bc5db hifigan config update 2021-04-08 11:20:33 +02:00
Eren Gölge 8d4fd79cd7 update hifigan config 2021-04-08 11:20:33 +02:00
rishikksh20 e656e8b108 Remove select size bug 2021-04-08 11:20:33 +02:00
rishikksh20 b533474e3b Remove minor bugs and make code trainable 2021-04-08 11:20:33 +02:00
rishikksh20 ef6ff4e95c Add Exponential LR scheduler check 2021-04-08 11:20:33 +02:00
rishikksh20 1535777f64 1) Add ExponentialLR 2021-04-08 11:18:36 +02:00
rishikksh20 c20a6b1185 * Format the model definition 
* Update code and integrate training code
 2021-04-08 11:18:36 +02:00
rishikksh20 39b5845810 1) Add hifigan json files 
2) Rename MPD disc
3) Re-format remove weight norm generator
 2021-04-08 11:14:39 +02:00
rishikksh20 7b7c5d635f 1) Combine MSD with Multi-Period disc 
2) Add remove weight norm layer on Generator
 2021-04-08 11:14:39 +02:00
rishikksh20 4493feb95c Add HiFi-GAN v1 generator and discriminator classes 2021-04-08 11:14:39 +02:00
Eren Gölge c86c559349 docstring and optional padding in TorchSTFT 2021-04-07 12:36:15 +02:00
Eren Gölge f890454de3 linter fixes 2021-04-07 12:36:03 +02:00
Eren Gölge 9782d9ea5d [ci skip] implement #418 2021-04-06 16:24:50 +02:00
Eren Gölge f46a275b22 update docstring 2 2021-04-06 16:24:50 +02:00
Eren Gölge ec94ff3691 update docstring 2021-04-06 16:24:50 +02:00
Eren Gölge 2048095e9a audio.py fix 2021-04-06 16:24:50 +02:00
Eren Gölge e0b3008c31 allow choosing the log function used for amptodb conversion 2021-04-06 16:24:50 +02:00
Eren Gölge 44b4cb5ba5 DCA comment 2021-04-06 16:24:50 +02:00
Eren Gölge b86e7fb2e8 pad short samples when loading precomputed features in vocoder trainign 2021-04-06 16:24:50 +02:00
Eren Gölge 6ad4eba678 gan vocoder train fix in case of restoring models wiht no scheduler is defined 2021-04-06 16:24:50 +02:00
Eren Gölge e3ccfe37ea add DE more urls 2021-04-02 14:54:41 +02:00
Eren Gölge e84f120a04 sam-accenture model preprocessor 2021-04-01 03:41:41 +02:00