A client asked if there's a way to extract more than one voice in a mix. His track has a male and a female singer and he needs JUST the female one. Of course today is easy to separate channels on a ready mix, but is there a way to separate more than one vocal on the same track?

 

interesting concept, but I don't think AI can achieve that yet. You will have already noticed that the algorithms struggle when there is voice and trumpet, or voice and guitar, as far as I know it is impossible to separate two different singers into separate tracks.

 

I had good results doing that with Melodyne as long as the two voices aren't happening at the same time (i.e duet singers one octave from each other).

Like this (with one voice):

 

Last edited: Nov 30, 2023

You will not get it perfect, but you could duplicate and frequency split the file first in the scenario of 1 female and 1 male singer in different octaves. EQ and other audio tools to get it closer before processing it. Then AI extract. It would be harder but I think you could probably do it to a somewhat acceptable result.