There is software like Spleeter [1] that can split songs into stems (seperate drums, vocals, rhythm, etc.) using ML. I imagine it would be possible to adapt it

There is software like Spleeter [1] that can split songs into stems (seperate drums, vocals, rhythm, etc.) using ML. I imagine it would be possible to adapt it for this. Might even work as-is if you only needed to isolate people's voices. 1. https://github.com/deezer/spleeter https://github.com/deezer/spleeter