It’s worth trying to run your files through Elevenlabs. They have some pretty good speech extraction/audio clean-up tools - it’s browser based, so it might not be anything near what you need.
it’s just impossible to do in a clean way. Too many overlapping frequencies which no algorithm can differentiate or separate. If you were a “top producer” and mixed everything with clarity and space or created everything in the box it would be easier but thank god you’re neither and you make cool tunes with hardwear so it’s glued on the way in already and it’s all mushed in the stereo field so you cannot unpick that. You’ll end up with whiney weasely stems that don’t really resemble what they would have been. I don’t think software or ai will ever be able to rip things perfectly, even if it learns what the sound might have been if it were stems, it’s still only guessing.
but it’s already started a new art movement right there! Surreal Morphic Realism.
OpenAI whistleblower found dead in San Francisco apartment
Nothing suspicious whatsoever there.