Commit Graph

1012 Commits

Author SHA1 Message Date
James Betker
8d01f7685c Get rid of absolute positional embeddings in unifiedvoice 2021-12-26 00:10:24 -07:00
James Betker
6700f8851d moar verbosity 2021-12-25 23:23:21 -07:00
James Betker
8acf3b3097 Better dimensional asserting 2021-12-25 23:18:25 -07:00
James Betker
e959541494 Add position embeddings back into unified_voice
I think this may be the solution behind the days problems.
2021-12-25 23:10:56 -07:00
James Betker
ab9cafa572 Make tokenization configs more configurable 2021-12-25 12:17:50 -07:00
James Betker
52410fd9d9 256-bpe tokenizer 2021-12-25 08:52:08 -07:00
James Betker
8e26400ce2 Add inference for unified gpt 2021-12-24 13:27:06 -07:00
James Betker
8b19c37409 UnifiedGptVoice! 2021-12-23 15:20:26 -07:00
James Betker
e55d949855 GrandConjoinedDataset 2021-12-23 14:32:33 -07:00
James Betker
c737632eae Train and use a bespoke tokenizer 2021-12-22 15:06:14 -07:00
James Betker
66bc60aeff Re-add start_text_token 2021-12-22 14:10:35 -07:00
James Betker
a9629f7022 Try out using the GPT tokenizer rather than nv_tacotron
This results in a significant compression of the text domain, I'm curious what the
effect on speech quality will be.
2021-12-22 14:03:18 -07:00
James Betker
7ae7d423af VoiceCLIP model 2021-12-22 13:44:11 -07:00
James Betker
09f7f3e615 Remove obsolete lucidrains DALLE stuff, re-create it in a dedicated folder 2021-12-22 13:44:02 -07:00
James Betker
a42b94ab72 gpt_tts_hf inference fixes 2021-12-22 13:22:15 -07:00
James Betker
48e3ee9a5b Shuffle conditioning inputs along the positional axis to reduce fitting on prosody and other positional information
The mels should still retain some short-range positional information the model can use
for tone and frequencies, for example.
2021-12-20 19:05:56 -07:00
James Betker
53858b2055 Fix gpt_tts_hf inference 2021-12-20 17:45:26 -07:00
James Betker
712d746e9b gpt_tts: format conditioning inputs more for contextual voice clues and less for prosidy
also support single conditional inputs
2021-12-19 17:42:29 -07:00
James Betker
c813befd53 Remove dedicated positioning embeddings 2021-12-19 09:01:31 -07:00
James Betker
b4ddcd7111 More inference improvements 2021-12-19 09:01:19 -07:00
James Betker
f9c45d70f0 Fix mel terminator 2021-12-18 17:18:06 -07:00
James Betker
937045cb63 Fixes 2021-12-18 16:45:38 -07:00
James Betker
9b9f7ea61b GptTtsHf: Make the input/target placement easier to reason about 2021-12-17 10:24:14 -07:00
James Betker
2fb4213a3e More lossy fixes 2021-12-17 10:01:42 -07:00
James Betker
9e8a9bf6ca Various fixes to gpt_tts_hf 2021-12-16 23:28:44 -07:00
James Betker
62c8ed9a29 move speech utils 2021-12-16 20:47:37 -07:00
James Betker
4f8c4d130c gpt_tts_hf: pad mel tokens with an <end_of_sequence> token. 2021-12-12 20:04:50 -07:00
James Betker
8917c02a4d gpt_tts_hf inference first pass 2021-12-12 19:51:44 -07:00
James Betker
5a664aa56e misc 2021-12-11 08:17:26 -07:00
James Betker
6ccff3f49f Record codes more often 2021-12-07 09:22:45 -07:00
James Betker
d0b2f931bf Add feature to diffusion vocoder where the spectrogram conditioning layers can be re-trained apart from the rest of the model 2021-12-07 09:22:30 -07:00
James Betker
662920bde3 Log codes when simply fetching codebook_indices 2021-12-06 09:21:43 -07:00
James Betker
380a5d5475 gdi.. 2021-12-03 08:53:09 -07:00
James Betker
101a01f744 Fix dvae codes issue 2021-12-02 23:28:36 -07:00
James Betker
07b0124712 GptTtsHf! 2021-12-02 21:48:42 -07:00
James Betker
85542ec547 One last fix for gpt_asr_hf2 2021-12-02 21:19:28 -07:00
James Betker
04454ee63a Add evaluation logic for gpt_asr_hf2 2021-12-02 21:04:36 -07:00
James Betker
5956eb757c ffffff 2021-11-24 00:19:47 -07:00
James Betker
f1ed0588e3 another fix 2021-11-24 00:11:21 -07:00
James Betker
7a3c4a4fc6 Fix lr quantizer decode 2021-11-24 00:01:26 -07:00
James Betker
3f6ecfe0db q fix 2021-11-23 23:50:27 -07:00
James Betker
d9747fe623 Integrate with lr_quantizer 2021-11-23 19:48:22 -07:00
James Betker
82d0e7720e Add choke to lucidrains_dvae 2021-11-23 18:53:37 -07:00
James Betker
934395d4b8 A few fixes for gpt_asr_hf2 2021-11-23 09:29:29 -07:00
James Betker
01e635168b whoops 2021-11-22 17:24:13 -07:00
James Betker
973f47c525 misc nonfunctional 2021-11-22 17:16:39 -07:00
James Betker
3125ca38f5 Further wandb logs 2021-11-22 16:40:19 -07:00
James Betker
0604060580 Finish up mods for next version of GptAsrHf 2021-11-20 21:33:49 -07:00
James Betker
14f3155ec4 misc 2021-11-20 17:45:14 -07:00
James Betker
555b7e52ad Add rev2 of GptAsrHf 2021-11-18 20:02:24 -07:00
James Betker
1287915f3c Fix dvae test failure 2021-11-18 00:58:36 -07:00
James Betker
019acfa4c5 Allow flat dvae 2021-11-18 00:53:42 -07:00
James Betker
f3db41f125 Fix code logging 2021-11-18 00:34:37 -07:00
James Betker
79367f753d Fix error & add nonfinite warning 2021-11-09 23:58:41 -07:00
James Betker
c584320cf3 Fix gpt_asr_hf distillation 2021-11-07 21:53:21 -07:00
James Betker
a367ea3fda Add script for computing attention for gpt_asr 2021-11-07 18:42:06 -07:00
James Betker
756b4dad09 Working gpt_asr_hf inference - and it's a beast! 2021-11-06 21:47:15 -06:00
James Betker
596a62fe01 Apply fix to gpt_asr_hf and prep it for inference
Fix is that we were predicting two characters in advance, not next character
2021-11-04 10:09:24 -06:00
James Betker
993bd52d42 Add spec_augment injector 2021-11-01 18:43:11 -06:00
James Betker
4cff774b0e Reduce complexity of the encoder for gpt_asr_hf 2021-11-01 17:02:28 -06:00
James Betker
da55ca0438 gpt_asr using the huggingfaces transformer 2021-11-01 17:00:22 -06:00
James Betker
83cccef9d8 Condition on full signal 2021-10-30 19:58:34 -06:00
James Betker
df45a9dec2 Fix inference mode for lucidrains_gpt 2021-10-30 16:59:18 -06:00
James Betker
92fe8b4dd9 ffffpt2 2021-10-29 17:29:49 -06:00
James Betker
95ca88efce Fix feedforward 2021-10-29 17:27:51 -06:00
James Betker
b476516340 Check in backing changes (which may have broken something?) 2021-10-29 17:22:33 -06:00
James Betker
986fc9628d Check in GPT with new inference methods (but not the backing code..) 2021-10-29 17:21:40 -06:00
James Betker
58494b0888 Add support for distilling gpt_asr 2021-10-27 13:10:07 -06:00
James Betker
5d714bc566 Add deepspeech model and support for decoding with it 2021-10-27 13:09:46 -06:00
James Betker
3a9d1c53ea Rework conditioning inputs provided 2021-10-26 10:46:33 -06:00
James Betker
43e389aac6 Add time_embed_dim_multiplier 2021-10-26 08:55:55 -06:00
James Betker
ba6e46c02a Further simplify diffusion_vocoder and make noise_surfer work 2021-10-26 08:54:30 -06:00
James Betker
0ee1c67ce5 Rework how conditioning inputs are applied to DiffusionVocoder 2021-10-24 09:08:58 -06:00
James Betker
06ea6191a9 Initial implementation of audio_with_noise dataset 2021-10-21 16:45:19 -06:00
James Betker
0dee15f875 base DVAE & vector_quantizer 2021-10-20 21:19:38 -06:00
James Betker
f2a31702b5 Clean stuff up, move more things into arch_util 2021-10-20 21:19:25 -06:00
James Betker
a6f0f854b9 Fix codes when inferring from dvae 2021-10-17 22:51:17 -06:00
James Betker
d016a2fbad Go back to vanilla flavor of diffusion 2021-10-17 17:32:46 -06:00
James Betker
23da073037 Norm decoder outputs now 2021-10-16 09:07:10 -06:00
James Betker
0edc98f6c4 Throw out the idea of conditioning on discrete codes. Oh well :( 2021-10-16 09:02:01 -06:00
James Betker
62c8c5d93e Zero out spectrogram code inputs initially. 2021-10-15 12:10:11 -06:00
James Betker
1d0b44ebc2 More tweaks to diffusion-vocoder 2021-10-15 11:51:17 -06:00
James Betker
3b19581f9a Allow num_resblocks to specified per-level 2021-10-14 11:26:04 -06:00
James Betker
83798887a8 Mods to support unet diffusion vocoder with conditioning 2021-10-13 21:23:18 -06:00
James Betker
33120cb35c Add norming to discretization_loss 2021-10-06 17:10:50 -06:00
James Betker
f2977d360c Allow attention_dim in channel attention to be specified, add converter 2021-10-05 17:29:38 -06:00
James Betker
9c0d7288ea Discretization loss attempt 2021-10-04 20:59:21 -06:00
James Betker
66f99a159c Rev2 2021-10-03 15:20:50 -06:00
James Betker
09f373e3b1 Add dvae with channel attention 2021-10-03 10:52:01 -06:00
James Betker
0396a9d2ca Increase baseline codes recording across all dvae models 2021-09-30 08:09:07 -06:00
James Betker
f84ccbdfb2 Fix quantizer with balancing_heuristic 2021-09-29 14:46:05 -06:00
James Betker
4914c526dc More cleanup 2021-09-29 14:24:49 -06:00
James Betker
6e550edfe3 Attentive dvae 2021-09-29 14:17:29 -06:00
James Betker
55b58fb67f Clean up codebase
Remove stuff that I'm likely not going to use again (or generally failed experiments)
2021-09-29 09:21:44 -06:00
James Betker
4d1a42e944 Add switchnorm to gumbel_quantizer 2021-09-24 18:49:25 -06:00
James Betker
ac57cdc794 Add scheduling to quantizer, enable cudnn_benchmarking to be disabled 2021-09-24 17:01:36 -06:00
James Betker
3e64e847c2 Gumbel quantizer 2021-09-23 23:32:03 -06:00
James Betker
c5297ccec6 Add dvae balancing heuristic 2021-09-23 21:19:36 -06:00
James Betker
e24c619387 Fix 2021-09-23 16:07:58 -06:00
James Betker
6833048bf7 Alterations to diffusion_dvae so it can be used directly on spectrograms 2021-09-23 15:56:25 -06:00
James Betker
5c8d266d4f chk 2021-09-17 09:15:36 -06:00
James Betker
a6544f1684 More checkpointing fixes 2021-09-16 23:12:43 -06:00
James Betker
94899d88f3 Fix overuse of checkpointing 2021-09-16 23:00:28 -06:00
James Betker
f78ce9d924 Get diffusion_dvae ready for prime time! 2021-09-16 22:43:10 -06:00
James Betker
6f48674647 Support diffusion models with extra return values & inference in diffusion_dvae 2021-09-16 10:53:46 -06:00
James Betker
0382660159 Get diffusion_dvae functional 2021-09-14 17:43:31 -06:00
James Betker
76e2c497f7 Improvements to splitter 2021-09-09 23:34:56 -06:00
James Betker
742f9b4010 Batch spleeter cleaner using GPU 2021-09-09 23:14:32 -06:00
James Betker
73b930c0f6 Add diffusion_dvae
Increase split_on_silence interval
2021-09-09 16:22:05 -06:00
James Betker
b8f2e0f452 mydvae 2021-09-06 17:45:30 -06:00
James Betker
3e073cff85 Set kernel_size in diffusion_vocoder 2021-09-01 08:33:46 -06:00
James Betker
dabd87246d Add unet_diffusion_vocoder 2021-08-31 14:38:33 -06:00
James Betker
909754cc27 Add find_faulty_files.py 2021-08-25 18:00:43 -06:00
James Betker
08b33c8e3a Support silu activation 2021-08-25 09:03:14 -06:00
James Betker
67bf7f5219 dvae mods
Trying to squeeze as much performance out of this net as possible
2021-08-25 08:55:13 -06:00
James Betker
b521d94b01 Make gpt-asr more configurable 2021-08-19 16:33:41 -06:00
James Betker
570ed327ed Stop dataset - attempt #2 2021-08-18 18:29:38 -06:00
James Betker
17453ccbe8 Revert mods to lrdvae
They didn't really change anything
2021-08-17 09:09:29 -06:00
James Betker
8332923f5c Two more tools to test the audio segmentor 2021-08-17 09:09:11 -06:00
James Betker
1fede41b7b Audio segmentor 2021-08-16 22:51:53 -06:00
James Betker
729c1fd5a9 Fix up max lengths to save memory 2021-08-15 21:29:28 -06:00
James Betker
9e47e64d5a Add gpt_segmentor model
The idea is to specifically train a model that extracts phrases from
audio clips.
2021-08-15 21:23:07 -06:00
James Betker
a826d5f658 Mods to dvae
- Add resblock to each layer
- Increase filter size for each layer
- Use SiLU
2021-08-15 20:54:10 -06:00
James Betker
b8bec22f1a Fix gpt_asr inference bug 2021-08-15 20:53:42 -06:00
James Betker
a523c4f932 Auto-normalize wav files by data type 2021-08-15 09:09:51 -06:00
James Betker
98057b6516 Make lrdvae use quantized mode in eval() 2021-08-14 23:43:01 -06:00
James Betker
ad3391bd96 Fix nan issue when interpolating audio 2021-08-14 20:42:01 -06:00
James Betker
d6a73acaed Allow processing of multiple audio sources at once from nv_tacotron_dataset 2021-08-14 16:04:05 -06:00
James Betker
007976082b GPT_asr for inference 2021-08-14 14:37:17 -06:00
James Betker
e1bdd3f7c7 Fix gpt_asr bug. Initial implementation of beam search 2021-08-13 22:47:00 -06:00
James Betker
cdee31c60b GPT_ASR 2021-08-13 15:02:18 -06:00
James Betker
f5a9b88ef6 tacotron cleaners: remove quotation marks
these don't really have relevance for tts or asr
2021-08-11 16:18:44 -06:00
James Betker
20586a8edc Fix LRDVAE bug with quantizer integration 2021-08-11 16:17:22 -06:00
James Betker
82fc69abfa Add "pure" evaluator
Which simply computes the training loss against an eval dataset
2021-08-09 14:58:35 -06:00
James Betker
080bea2f19 No, really 2021-08-09 12:02:31 -06:00
James Betker
e1ce4671e4 Apply dropout to gpt_tts, get rid of min_gpt implementation 2021-08-09 12:01:10 -06:00
James Betker
1068f53b78 Add a sampling beam search 2021-08-09 11:56:06 -06:00
James Betker
01cfae28d8 Beam search implementation in one pass? Dayyyum 2021-08-08 23:22:42 -06:00
James Betker
690d7e86d3 Fix nv_tacotron_dataset bug which incorrectly mapped filenames
dammit..
2021-08-08 11:38:52 -06:00
James Betker
a2afb25e42 Fix inference, always flow full text tokens through transformer 2021-08-07 20:11:10 -06:00
James Betker
4c678172d6 ugh 2021-08-06 22:10:18 -06:00
James Betker
e723137273 Make gpttts more configurable 2021-08-06 22:08:51 -06:00
James Betker
a7496b661c combined dvae ftw 2021-08-06 22:01:06 -06:00
James Betker
0237e96b34 Fix dvae bug 2021-08-06 14:17:01 -06:00
James Betker
0799d95af5 Use quantizer from rosinality/vqvae with openai dvae 2021-08-06 14:06:26 -06:00
James Betker
d3ace153af Add logic for performing inference using gpt_tts with dual-encoder modes 2021-08-06 12:04:12 -06:00
James Betker
b43683b772 Add lucidrains_dvae 2021-08-06 12:03:46 -06:00
James Betker
70dcd1107f Fix byol_model_wrapper to function with audio inputs 2021-08-05 22:20:22 -06:00
James Betker
89d15c9e74 Move gpt-tts back to lucidrains implementation
Much better performance.
2021-08-05 22:15:13 -06:00
James Betker
d120e1aa99 Add audio augmentation to wavfile_dataset, utility to test audio similary 2021-08-05 22:14:49 -06:00