# Extracting Voice

heyas, i am brand new to this sort of thing…

I was just wondering how you would extract the vocals of the song so you could add a diff beat to it?

Thanks

Al

99.99% of the time you can’t. It would be like taking the chocolate out of a cake that’s already been baked.

If you had an exact instrumental copy of a song and a copy with vocals you could subtract the one from the other and be left with only vocals. You could also do it if all the vocals were in one channel by themselves and all the instruments were in the other channel.

Also, once in a great while you can use Noise Removal to make the vocals easier to understand, but they will sound terrible. If the song has a very constant backing track and the voice sounds much different from the rest of it, you can select an instrument only section as the Noise Profile and then apply the Noise Removal to the whole thing. But you’ll never get workable results out of this method.

delete

Chris,

Actually, the theory isn’t on your side here. I don’t think this kind of mixing can be done by adding / subtracting. It might theoretically be possible to write some code that would compare the left and right tracks and remove anything that is the same within them in order to get a vocal-less track that you could then use to remove the music with. But I’ve never seen one.

In other words, right now you have a mono track that is either (Left - Right) OR (Right - Left). And you’re going to try to subtract that from a stereo signal (Left + Right).

The best you can do working that way is get Left OR Right back (at twice the volume), you can’t use that to pull the center out since it depends on both Left and Right.

Mathematically, all you can do is this:

(Left - Right) + (Left + Right) = Left * 2

(Right - Left) + (Left + Right) = Right * 2

Does that make sense?

Hey, about this… ive been messn with it, but could you maybe use equalization and remove the lower frequencies, to isolate the higher pitch of the voice? i tried, but you could still slightly hear the BGM, which was annoying. could you maybe develop that code that you were talking about? im not smart enough to do a complex code like that.

Unfortunately that does not quite work.
If we consider a stereo recording which has a vocal in the centre, instrument A on the left and instrument B on the right.

On the Left channel we have instrument A + half of the Vocal signal
We could write this as Lc = A + V/2

On the Right channel we have instrument B + half of the Vocal
We could write this as Rc = B + V/2

Now putting this into your formula “(Left - Right) + (Left + Right)” we have
(Lc -Rc) + (Lc + Rc) = ((A+V/2) - (B+V/2)) + ((A+V/2) + (B+V/2))
=(A-B) + (A+B+V)
=2A+V

similarly with “(Right - Left) + (Left + Right)” we end up with 2B+V

So you will see that we never manage to get rid of the centre vocal.

Using the same logic, we can see the usual “Karaoke” method is:
Lc - Rc = (A+V/2) - (B+V/2)
=A-B

Uhhh, Steve:

Now putting this into your formula “(Left - Right) + (Left + Right)” we have
(Lc -Rc) + (Lc + Rc) = ((A+V/2) - (B+V/2)) + ((A+V/2) + (B+V/2))
=(A-B) + (A+B+V)
=2A+V

Yes, I agree with that. But look at how you’ve defined the Left channel:

Lc = A + V/2

That’s exactly equal to what I said it would be. Left * 2.

So yes, we can’t use that method to erase the edges of the stereo spectrum, only the middle of it.