I’ve been focusing quite a bit on this general problem domain, and have a picked up a few things that might be useful – though my interest is in isolating the vocals, not removing them – but a lot of what I’ve found is probably still interesting and helpful here.
First, here is a free open source package that works in a way that is similar to Melodyne, although it has a very different workflow and only focuses on isolating frequency bands – that is what Melodyne does, but it has a lot more complicated stuff involved that makes the process more powerful… but free is a very good price.
ISSE
I also found this really great pair of videos from DJ Vespers that covers how to do pretty effective vocal isolation in Ableton Live – this is pretty powerful. Imperfect, as all mixed audio separation tends to be, but with enough work, you can really achieve some great stuff.
How to make acapellas in Ableton Live HD tutorial pt 1: mid side EQ Eight
Once you have isolated the vocal track, you can then use phase inversion to remove it from the mix and arrive at your instrumental.
Research in this field is pretty impressive, I linked an article elsewhere on this forum that related to this, and have been reading a lot of other academic papers on the subject. I am not sure whether I can fit developing my own tool into my free time, but it is something that I’d really like to do because the solutions I am seeing from the academic world seem to be quite a lot further along than what is currently available commercially. It is theoretically possible, without guidance, to isolate each instrument into its own channel, which is very exciting stuff.
.
