Comment by Hydraulix989

It's possible, but the quality isn't that great even with state of the art machine learning:

https://towardsdatascience.com/audio-ai-isolating-vocals-fro...