Commits · a18a0cabe6d52a2050489b99e55e9a4f7624ee92 · 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 17, 2022
- another fix needed for variable lengthed coarse token ids · a18a0cab
  Phil Wang authored Nov 16, 2022
  
  View commits for tag 0.0.39 0.0.39
  
  a18a0cab
- cleanup · a7cb08d8
  Phil Wang authored Nov 16, 2022
  
  View commits for tag 0.0.38 0.0.38
  
  a7cb08d8
- prepare for variable lengthed coarse tokens in fine transformer as well 0.0.36 · bc626a64
  Phil Wang authored Nov 16, 2022
  
  View commits for tag 0.0.36 0.0.36
  
  bc626a64
- just give coarse sequence in coarse transformer its own start token · 2c2d43d8
  Phil Wang authored Nov 16, 2022
  
  View commits for tag 0.0.34 0.0.34
  
  2c2d43d8
- ensure semantic transformer generations always include an eos for each sequence · 5cf4e945
  Phil Wang authored Nov 16, 2022
  
  5cf4e945
- semantic token ids will have variable lengths because of unique consecutive,... · 70f02c5b
  Phil Wang authored Nov 16, 2022
  70f02c5b
- cleanup · c6bcd118
  Phil Wang authored Nov 16, 2022
  
  c6bcd118
- handle unique consecutive issue with generating semantic token ids · ea194173
  Phil Wang authored Nov 16, 2022
  
  View commits for tag 0.0.31 0.0.31
  
  ea194173
- only encode raw text once when generating · 0806f62b
  Phil Wang authored Nov 16, 2022
  
  View commits for tag 0.0.30 0.0.30
  
  0806f62b
- code for sampling semantic token ids from semantic transformer, input can be... · d9152df6
  Phil Wang authored Nov 16, 2022
  View commits for tag 0.0.29 0.0.29
  
  d9152df6
- prepare some sampling functions · 3ea0bcf4
  Phil Wang authored Nov 16, 2022
  
  View commits for tag 0.0.28 0.0.28
  
  3ea0bcf4
- get some working scripts into readme · 6a0c7e73
  Phil Wang authored Nov 16, 2022
  
  View commits for tag 0.0.27 0.0.27
  
  6a0c7e73
- move soundstream to separate file · 416aaf7a
  Phil Wang authored Nov 16, 2022
  
  416aaf7a
Nov 16, 2022
- readme · f9fa68e7
  Phil Wang authored Nov 15, 2022
  
  f9fa68e7
- readme · 873f3c2e
  Phil Wang authored Nov 15, 2022
  
  873f3c2e
- oops · 95b8cd23
  Phil Wang authored Nov 15, 2022
  
  View commits for tag 0.0.25 0.0.25
  
  95b8cd23
- optionally allow for resampling directly within SoundDataset, if target_sample_khz specified · b6e5af78
  Phil Wang authored Nov 15, 2022
  
  b6e5af78
- a simple measure for greater transformer training stability · f7756f56
  Phil Wang authored Nov 15, 2022
  
  View commits for tag 0.0.23 0.0.23
  
  f7756f56
- handle if any of the models requires the sequence length to be some multiple of · 5b24b4f5
  Phil Wang authored Nov 15, 2022
  
  View commits for tag 0.0.22 0.0.22
  
  5b24b4f5
- make sure unconditional synthesis can still work, add ability to resample... · 02902731
  Phil Wang authored Nov 15, 2022
  View commits for tag 0.0.21 0.0.21
  
  02902731
- fix a bug thanks to @eonglints · 2725ae89
  Phil Wang authored Nov 15, 2022
  
  View commits for tag 0.0.20 0.0.20
  
  2725ae89
Nov 15, 2022
- properly accelerator prepare all multiscale discriminators · 5f964966
  Phil Wang authored Nov 14, 2022
  
  View commits for tag 0.0.19 0.0.19
  
  5f964966
- take care of training all multiscale discriminators · 4a206d7b
  Phil Wang authored Nov 14, 2022
  
  View commits for tag 0.0.18 0.0.18
  
  4a206d7b
- 0.0.17 · 07a7eb00
  Phil Wang authored Nov 14, 2022
  
  View commits for tag 0.0.17 0.0.17
  
  07a7eb00
- get some training code down for soundstream, use torchaudio instead of soundfile · a5c3ace4
  Phil Wang authored Nov 14, 2022
  
  a5c3ace4
- oops · bc2e9461
  Phil Wang authored Nov 14, 2022
  
  View commits for tag 0.0.16 0.0.16
  
  bc2e9461
- fix a bug with residual quantize dropout, and also figure out a way to deal... · 38175d4b
  Phil Wang authored Nov 14, 2022
  View commits for tag 0.0.15 0.0.15
  
  38175d4b
- 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