Package request: Demucs

[KaiVinter]

• [x] I understand TUR package policy

Describe package: Name: Demucs Music Source Separation Homepage: ↓ Source link: https://github.com/facebookresearch/demucs

Additional Info: This is a really useful tool for musicians! This is an AI based audio source separation tool. It's written in Python. I tried instaling it as the GitHub page says, but it gave me some errors and couldn't install. If this happened, is there a way to erase the files that were downloaded? Since it didn't install correctly via pip, there's no need to keep the Demucs files of a failed installation

[licy183]

It seems that this program needs cuda. But cuda is impossible to work on Termux...

https://github.com/facebookresearch/demucs/blob/bd98f01aac5a9e1b1db55cb5744c1ffa69ff39e1/requirements_minimal.txt#L1

[nacho00112]

• [x] I understand TUR package policy

Describe package: Name: Demucs Music Source Separation Homepage: ↓ Source link: https://github.com/facebookresearch/demucs

Additional Info: This is a really useful tool for musicians! This is an AI based audio source separation tool. It's written in Python. I tried instaling it as the GitHub page says, but it gave me some errors and couldn't install. If this happened, is there a way to erase the files that were downloaded? Since it didn't install correctly via pip, there's no need to keep the Demucs files of a failed installation

If you used pip then do this:

pip cache purge

Sometimes there also are files in $PREFIX/tmp that are not deleted if pip is interrupted in a bad way so:

# make sure that you not have anything important there
rm -r $PREFIX/tmp/*

[knyipab]

I think it's possible. I installed demucs with these commands on my devices:

pkg install -y python-lameenc python-torchaudio
pip install demucs

And then test with:

demucs input_audio.mp4 -n hdemucs_mmi -d cpu --mp3 --two-stems=vocals

It outputs separated audio in separated dir.

Last words: still, it is 7~9 times slower than spleeter-proot (#1037) on my devices.