Another bit of free software for the sample heads here -
Some of the details about this project are that they implement various AI models to do the separation, the best of which (to my ears) are the Demucs ones which can do up to 6 stems. And Demucs are based off of open source models from Facebook labs (thank you Meta).
You need a few gigs of hard disk space but well worth it if your interested in this sort thing, they also chuck in a time stretch algo that is pretty f’ing good.
I set up on windows 11 with zero effort/issues, their downloadable contains all the dependencies needed incl Python, torch etc.
And it runs very fast if you have an nvidia cuda enabled GPU.
What a time to be alive 
and making music.