search

lucidrains / audiolm-pytorch

Implementation of AudioLM, a SOTA Language Modeling Approach to Audio Generation out of Google Research, in Pytorch
MIT License
2.39k stars 255 forks source link

Poor audio quality #191

Closed cpdu closed 1 year ago

cpdu commented 1 year ago

Hi,

Thank you so much for sharing the codes. We try to train an AudioLM model on LibriTTS but we find it has a poor quality and is much worse than the official demos.

Do you have any comments on that? I'm wondering that have you ever achieved a similar quality to the official demos in your training?

lucidrains commented 1 year ago

@cpdu normally i would dig into it, but i've already received multiple reports of people training successfully (in discussions as well as through my email)

if you need help with the training, recommend opening up a discussion if someone is charitable enough to handhold you through it

you will need to elaborate more than simply "it is not working". training plots in weights and biases, hyperparameter, dataset etc