[Feature Request] Use demucs for audio enhancement! - Githubissues

Vaibhavs10 / insanely-fast-whisper

Apache License 2.0

7.68k stars 537 forks source link

[Feature Request] Use demucs for audio enhancement! #101

Open Vaibhavs10 opened 11 months ago

Vaibhavs10 commented 11 months ago

It should be an easy addition. It'll help with use-cases where audio is too garbled.

API looks exceptionally clean: https://github.com/facebookresearch/demucs/blob/main/docs/api.md It also works on Mac! 👍

(Putting this up so that I don't forget to add this.)

gaspardpetit commented 9 months ago

Adding my 2 cents here - having played with both demucs and MDX-Net (https://github.com/karaokenerds/python-audio-separator) from Ultimate Vocal Remover for similar purposes, I found that demucs ended up adding interfering artifacts to the audio while MDX-Net produced cleaner output.

In case it helps:

A sample showing how I used demucs: https://github.com/gaspardpetit/verbatim/blob/main/verbatim/voice_isolation/isolate_voices_demucs.py
A sample showing how I used MDX-Net: https://github.com/gaspardpetit/verbatim/blob/main/verbatim/voice_isolation/isolate_voices_mdx.py