Interesting, I wonder why this couldn’t be accomplished with conventional techniques. I own a handful of “AI” Plugins meant to achieve similar cleanup and I feel like it always needs to be tweaked to sound right. And that’s for a guy like me without practiced mixing ears. I wonder why real studio engineers needed AI.

Yeah. From what I understand, Giles Martin used the same technique (or, at least, a similar one) when making the 2022 remix of Revolver. The AI was able to make clean stems from the original tracks where the instruments were mixed. So, I would suspect that this technique was able to separate the vocal from tape hiss, room noise and other factors–it isn’t a deepfake of John Lennon.