Isolating an Instrumental version

So a friend told me there was programs that could do that. I wonder how?

I did something vaguely similar before, by manually aligning loops in a track and just fondling with the volume of each layer until I had only the bits that contained no voice. But this was a real pain in the ass since I pretty much had to use the spaces where the singer stopped to breath, and it only works when there's plenty of identical loops with not-so-persistent singing.

I imagine a program could work somewhat like that, kinda like a subtractive noise reduction method, only without the quality loss assuming you got multiple sources of the same loop so the voice can be eliminated. But is there really one that does that?

phase inversion?

You can remove the sounds that are equally loud in both channels, which many times removes vocals.

1. Load up the track in a DAW or sound editing program.

2. Split the stereo file into two mono files.

3. Phase reverse one of the two sides.

4. Pan both sides to center.

I didn't know about that phase inversion, I checked how to do it and I'm pretty sure I pulled it off right, now it just sounds very odd, some elements are kind of clearer but altogether it sounds very glitchy and the bass is almost completely gone.

I fondled about with the panning but that doesn't do the trick either.

Maybe I should try it with another track? But I still don't get how it's supposed to work, aren't the vocals and the instruments on both channels like an indistinguishable mash?

There was a good article about stereo sound theory in SOS a couple months back.

Here:

http://soundonsound.com/sos/nov10/articles/stereoprocessing.htm

The joys of posting from the phone...

Just google "sound on sound stereo theory".

What phase inversion does is kill anything in the center of the stereo field. essentially you are subtracting one side from the other and you are left which whatever elements of the song are panned out, the harder they are panned, the more clear they will be in the result. While this often works well to kill vocals because lead vox are almost alwyas in the center of the stero field, you will also likely lose bass and kick or snare since these are rarely panned as well. It's a tricky enterprise but it works well on some songs, not so well on others though..

I tried four different songs, one Busdriver song, two Talib Kweli tracks and another one by Portishead. But they all sounded terrible. In the Busdriver one, I managed to get a piano sample very clearly, and further into the song, there was a section where the voice was clear and the rest was mostly gone.

Is there a chance I will be able to fix it with some additional settings, panning etc, or does the center panning already push the effect to its max?

I fear there's not gonna be any accapella versions of the tracks I'm trying to isolate. Mostly not very popular hip hop tracks (which aren't overly complex in structure, so I'm still considering that manual pain in the ass method)

But if there was an accapella version, it could be subtracted via that same method? Would it sound like the vocals were never there or just "acceptable"?

generally more than acceptable, sometimes "quite good"

Melodyne?

It's a shame I can't do programming like that. I imagine if someone actually got down to writing a script that does what I did manually, you would get very high quality results ...

jonas had the only real solution here, but he's right: you'll usually find instrumentals before you find acapellas.

I tried the acapella thing - merged an instrumental track with a prank call I recorded, then phase-inverted the original prank call and subtracted it. Worked very well, there's some clipping but otherwise, the quality of the music doesn't seem to be affected by it and the voice is completely gone.

That may not help me with my original problem since I won't get my hands on an acapella version of the tracks I want to instrumentalize, but it's cool to know, I didn't know it was possible.

Does subtractive noise reduction work by using a similar method? But then why does it tend to create the glitchy bird noises?