Commits · b99a260f36f58c8eb1bde4479abd8ebe4a634dc4 · school / Capstone Design / 01 / AudioLM

Browse files

This project is mirrored from https://github.com/lucidrains/audiolm-pytorch.git. Pull mirroring failed May 17, 2023.
Repository mirroring has been paused due to too many failed attempts. It can be resumed by a project maintainer or owner.
Last successful update May 15, 2023.

Nov 15, 2022
- basic dataset and dataloader for audio, tested with librispeech · b99a260f
  Phil Wang authored Nov 14, 2022
  
  b99a260f
- will be using soundfile and torchaudio · dab7d1b8
  Phil Wang authored Nov 14, 2022
  
  dab7d1b8
Nov 12, 2022
- fix coarse cross entropy loss weights · 450af495
  Phil Wang authored Nov 11, 2022
  
  450af495
- complete first pass at unique consecutive issue with semantic token ids, by... · 09c79a04
  Phil Wang authored Nov 11, 2022
  View commits for tag 0.0.12 0.0.12
  
  09c79a04
- correct weighting of cross entropy losses · e5408fbd
  Phil Wang authored Nov 11, 2022
  
  View commits for tag 0.0.11 0.0.11
  
  e5408fbd
- follow researcher @eonglints advice and add unique consecutive for semantic... · 2b6c5662
  Phil Wang authored Nov 11, 2022
  View commits for tag 0.0.10 0.0.10
  
  2b6c5662
Nov 11, 2022
- product management · ccf9c1d8
  Phil Wang authored Nov 10, 2022
  
  ccf9c1d8
- add classifier free guidance training logic, cite · 30f04de7
  Phil Wang authored Nov 10, 2022
  
  View commits for tag 0.0.9 0.0.9
  
  30f04de7
- add cross attention layers as well as setup t5 and some conditioning logic,... · c17ee7d3
  Phil Wang authored Nov 10, 2022
  View commits for tag 0.0.8 0.0.8
  
  c17ee7d3
- gratitude · 26dfc80f
  Phil Wang authored Nov 10, 2022
  
  26dfc80f
- go for single-headed key / values for all decoding attention networks, given... · fca12286
  Phil Wang authored Nov 10, 2022
  View commits for tag 0.0.7 0.0.7
  
  fca12286
- listen to @eonglints and add hubert with kmeans as an option · a11722e6
  Phil Wang authored Nov 10, 2022
  
  View commits for tag 0.0.6 0.0.6
  
  a11722e6
Nov 09, 2022
- product management · ed313d3a
  Phil Wang authored Nov 08, 2022
  
  ed313d3a
Nov 08, 2022
- product management · fb752a74
  Phil Wang authored Nov 07, 2022
  
  fb752a74
- add an adapter class for fairseq vq-wav2vec, make sure training of semantic... · af0564d4
  Phil Wang authored Nov 07, 2022
  View commits for tag 0.0.5 0.0.5
  
  af0564d4
- will be depending on fairseq vq-wav2vec implementation... · 2ce09315
  Phil Wang authored Nov 07, 2022
  2ce09315
Nov 05, 2022
- optional · 8f5d07d3
  Phil Wang authored Nov 04, 2022
  
  8f5d07d3
- project management · 130846e3
  Phil Wang authored Nov 04, 2022
  
  130846e3
- offset by multiple of codebook size across quantizers for both coarse and fine · d40e4119
  Phil Wang authored Nov 04, 2022
  
  View commits for tag 0.0.4 0.0.4
  
  d40e4119
- rough sketch of all three transformers finished · 0ac5e4f5
  Phil Wang authored Nov 04, 2022
  
  0ac5e4f5
- some project management · 80a3fad4
  Phil Wang authored Nov 04, 2022
  
  80a3fad4
- credit assign · fa750085
  Phil Wang authored Nov 04, 2022
  
  fa750085
- complete semantic transformer, as it is a normal transformer · fac3152e
  Phil Wang authored Nov 04, 2022
  
  fac3152e
- handle projection of fine and coarse logits correctly in the final transformer in the hierarchy · 0ec7667b
  Phil Wang authored Nov 04, 2022
  
  0ec7667b
Nov 04, 2022
- todo · a9efd2d9
  Phil Wang authored Nov 03, 2022
  
  a9efd2d9
- always use shazeer glu feedforward · 95280273
  Phil Wang authored Nov 03, 2022
  
  95280273
- get some code for fine transformer (last stage) into place · 9c6785b8
  Phil Wang authored Nov 03, 2022
  
  9c6785b8
- update to latest vq library, make sure residual quantizer can dropout during... · 90904ee5
  Phil Wang authored Nov 03, 2022
  90904ee5
Nov 01, 2022
- will use gradient penalty on discriminator, as had much success with that in stylegan · 3ae3a0dd
  Phil Wang authored Oct 31, 2022
  
  3ae3a0dd
Oct 31, 2022
- Merge pull request #12 from shanguanma/main · 0a714012
  Phil Wang authored Oct 30, 2022
  0a714012
- fixed a errors of STFTDiscriminator · 74a9e42e
  shanguanma authored Oct 31, 2022
  
  74a9e42e
- residual vq ready for distributed training · 2a251506
  Phil Wang authored Oct 30, 2022
  
  View commits for tag 0.0.3 0.0.3
  
  2a251506
Oct 30, 2022
- use l1 loss for "feature" loss · 8910dbfe
  Phil Wang authored Oct 29, 2022
  
  View commits for tag 0.0.2 0.0.2
  
  8910dbfe
- Merge pull request #9 from aabzaliev/main · 532d8e95
  Phil Wang authored Oct 29, 2022
  532d8e95
- include stft discriminator features into feature loss · 42b334cf
  abzaliev authored Oct 29, 2022
  
  42b334cf
Oct 29, 2022
- add the "feature" loss as mentioned in the paper with customizable weight · d0e1f681
  Phil Wang authored Oct 28, 2022
  
  d0e1f681
Oct 28, 2022
- remove padding in causal convtranspose1d thanks to @NPN for catching that · 6810d515
  Phil Wang authored Oct 27, 2022
  
  6810d515
- first pass at stft discriminator · 6bd8b8fa
  Phil Wang authored Oct 27, 2022
  
  6bd8b8fa
- trims output on right for causal convtranspose1d, thanks to @NPN in... · d9c6ba40
  Phil Wang authored Oct 27, 2022
  d9c6ba40
Oct 27, 2022
- complete the multi-scale discriminators for soundstream · 4acee04e
  Phil Wang authored Oct 26, 2022
  
  4acee04e