I've just been informed that the magical new AI-based voice / accompaniment separation models are actually open-source instead of proprietary-only.