0.0.21

fix masking of upper quantizers during training, thanks to @chenht2010

0.0.20

end-to-end text-to-speech with audiolm, spear-tts, and soundstorm

0.0.19

remove the eos token id from the semantic conditioning

0.0.18

add ability to use t5 relative positional bias, addressing https://gi…

0.0.17

now able to train directly on raw audio, conditioned on semantic toke…

0.0.16

soundstorm can now accept semantic token ids for conditioning, which …

0.0.15

bump to latest audiolm

0.0.14

release correct masking strategy, thanks to @stevenhillis !

0.0.12

use a bias-less layernorm and remove batchnorm from conformer

0.0.11

add conformer, switch to rotary embeddings