Thought some people might be interested in this, the Algoriddim Djay app on iOS has added a “neural mix” feature which allows you to individually mix the beat/melody/vocals of any song, using “AI” source separation technology: https://cdm.link/2020/06/djay-pro-ai-stem-separation/
And VirtualDJ announced the same feature on the same day too, haha: https://djtechtools.com/2020/06/18/virtual-dj-2021-released-adds-dynamic-stems-signal-separation-mixing/
It seems to use the same kind of technology as Deezer’s Spleeter (https://deezer.io/releasing-spleeter-deezer-r-d-source-separation-engine-2b88985e797e?gi=6c747566a0f3) and Open-Unmix (https://sigsep.github.io/open-unmix/) and gets similar results – it’s absolutely not perfect, but I think it’s very impressive given what a seemingly impossible thing to do it is, and I think you could totally get away with using an acapella isolated in this way for a mash up or whatever.
Will be interesting to see how this technology progresses – machine learning has allowed massive leaps compared to the state of the art previously. Right now, I’d say the quality isn’t good enough for professional use aside from maybe the odd “mash up”, but being able to isolate individual parts of songs for DJing could be huge if they can continue to make such big leaps in quality.
There’s also a Max for Live version of Spleeter: https://azuki.bandcamp.com/merch/max-for-live-stem-splitter-spleeter and an experimental real-time VST version: https://azuki.bandcamp.com/merch/experimental-spleeter-plugin-for-live-stem-separation if you’re interested in having a play.