search

facebookresearch / demucs

Code for the paper Hybrid Spectrogram and Waveform Source Separation
MIT License
8.1k stars 1.01k forks source link

Separating one vocal from another #328

Open daleyguy opened 2 years ago

daleyguy commented 2 years ago

Hello. I've been using Demucs for about a year and love it. I've recently been wanting to separate/isolate two different vocals on a mono track. I did a search and found a page entitled "A Deep-Learning Based Framework for Source Separation, Analysis, and Synthesis of Choral Ensembles" that outlines attempts at this using various applications.

I've never attempted training of any kind, but would love to try it if it's feasible. I am able to save an example of one of the vocals when it sings a line all on it's own. That is the vocal I want to remove. Would it be possible to use that segment to train Demucs to extract the two vocals? Thanks.

(Please ask your question here.)

bendudz commented 2 years ago

I am too looking for a way to identify & separate individual singers from music. In my case I don't need the separate tracks, just to identify & label the main track. I found a paper titled Singer Diarization for Polyphonic Music With Unison Singing that may help you.