search

adefossez / demucs

Code for the paper Hybrid Spectrogram and Waveform Source Separation
MIT License
1k stars 96 forks source link

Update requirements.txt #7

Open oxysoft opened 4 months ago

oxysoft commented 4 months ago

Support latest version of torchaudio (already functional with no changes)

oxysoft commented 4 months ago

I had some problems using demucs in my project / cloud compute which uses torch 2.3.1, however I have updated all the libraries 2.3.1 and kept demucs and it's working just fine. Not all features have been tested, so keep that in mind

CarlGao4 commented 4 months ago

Actually, the global backend selector has been completely removed since torchaudio 2.2. Users can't switch audio backend when separating, so the separation won't be affected using torchaudio 2.2+. However, training requires this feature or passing the desire audio backend would make the code mess. Besides, deciding how to specify the backend according to torchaudio version will make the logic even complex, so I think limiting torchaudio versions to below 2.2 is a good choice.