Kn0ck0ut is a free VST plugin that was made to help extract vocals from music in digital audio, sometimes it works.
Kn0ck0ut takes two mono inputs, spectrally subtracts one from the other, and returns the difference. It can also retrieve centre-panned audio from stereo input.
Or get the Linux LV2 by J. Salwen here.
Installation (windows only): put the file Kn0ck0ut.DLL in your VST plugins directory. restart yr host app. set buffer size to at least 1024 samples unless you like the sound of clicks. add Kn0ck0ut as an insert effect on a stereo track or stereo bus.
If the ‘extract centre’ control is set to zero the plugin expects two mono inputs – one hard-panned left and one hard-panned right. The right channel signal is subtracted from the left signal the difference is returned on both channels. Example: if you have a vocal version and instrumental version of the same track, pass them thru Kn0ck0ut synced up best you can, the vocal version hard-panned left and the instrumental hard-panned right. The R will be subtracted from the L, maybe leaving just the vocals, most likely leaving plenty other stuff too.
Hints n tips:
Adjust input volumes on yr DAW for best results.
This not a phase-cancel process – audio doesn’t need to be in phase. But audio will clip if the signal peaks – use yr ears, watch the meters.
If you haven’t got the full instrumental try using an instrumental verse or chorus from the same track. Or record yr own instrumental version and use that.
Not only for vocals, you can recover whatever instrument or sound is the difference between L&R. Except drums, drums don’t sound good, get smeared.
For noise reduction, loop a sample of the noise and use that as the R channel input.
For dereverb, send yr too-reverby signal to the Kn0ck0ut L channel, send it also to a reverb patch with similar character and route the return (just verb with no dry signal) to the Kn0ck0ut R channel.
If ‘extract centre’ is set to one the plugin expects a stereo input and returns the centre-panned part of the stereo image on both channels. In some few and especially v old school stereo mixes the centre of the stereo image is just vox.