Skip to content
GitLab
Explore
Sign in
This is an archived project. Repository and other project resources are read-only.
Commits · 0dbb294d0280142c9b2fe3f758ae8b57f200e231
audiolm-pytorch-flask
Browse files
Nov 18, 2022
show import
· 0dbb294d
Phil Wang
authored
Nov 17, 2022
0dbb294d
allow for turning off cross entropy loss to preceding tokens (semantic,...
· 84bc71d7
Phil Wang
authored
Nov 17, 2022
View commits for tag 0.0.46
0.0.46
84bc71d7
reminder
· b0fbb485
Phil Wang
authored
Nov 17, 2022
b0fbb485
always train with quantize dropout in soundstream
· 490dbec1
Phil Wang
authored
Nov 17, 2022
View commits for tag 0.0.45
0.0.45
490dbec1
for the researchers
· fabccfd7
Phil Wang
authored
Nov 17, 2022
View commits for tag 0.0.44
0.0.44
fabccfd7
not sure how to handle, but make a guess
· acf0cff5
Phil Wang
authored
Nov 17, 2022
acf0cff5
use a hack to fix an issue, refactor semantic later
· 0b078214
Phil Wang
authored
Nov 17, 2022
View commits for tag 0.0.43
0.0.43
0b078214
make it work end to end
· 9c4c56d3
Phil Wang
authored
Nov 17, 2022
View commits for tag 0.0.42
0.0.42
9c4c56d3
be able to reconstruct the coarse wav from coarse transformer to soundstream decoder
· a17b189a
Phil Wang
authored
Nov 17, 2022
View commits for tag 0.0.41
0.0.41
a17b189a
add some preliminary generation code for coarse transformer, primed on semantic tokens
· 684da45c
Phil Wang
authored
Nov 17, 2022
View commits for tag 0.0.40
0.0.40
684da45c
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
Loading
