demucs process

[filmo]

Is there a reason that demucs.separate is called via a os.system() call rather than keeping it internal to the existing python process using the method described in the documentation??

import demucs.separate
demucs.separate.main(["--mp3", "--two-stems", "vocals", "-n", "mdx_extra", "track with space.mp3"])

My guess is it's just as a convenience as the current way was the first method listed in the documentation. I'm thinking of bringing it internal so that the scripts can respect CUDA_VISIBLE_DEVICES more easily and/or be able to pass a cuda device ID in so that all models and processes respect the selected cuda device.

[MahmoudAshraf97]

This API is fairly new and when it was released I felt like it's the same as calling a separate process, so I kept the separate process to guarantee that ram and vram will be cleared on finishing, but you raised some good points