This sounds like a great ml app idea: Remove music from audio while keeping the other voices. I wonder how hard it would be to keep them separate.
[1] https://github.com/deezer/spleeter/releases [2] https://paperswithcode.com/task/speaker-separation
I haven't tried speaker separation, but it seems there are some models from SpeechBrain on HuggingFace that seem to handle this: