in what sense? like just the vocals?
that would be like a combination of 'voice isolation software' and 'ms - voice recognition software'.
both of which are existing so you might wanna peg on that. that's a lot of algorithms to work on.
but that would be an awesome project.