Creating Accapella Versions

Hi, I am new to these forums so sorry if this is in the wrong place/answered before.
I use windows.

Recently on audacity I have been trying to create an Acapella for David Guetta’s Lovers on the Sum and after following instructions from many different sites ( amongst others) it all seems to come down to using the invert key on the karaoke track which is synced up but for some reason it does nothing. Yeah it flips the sound waves but it doesn’t seem to change anything sound wise - basically it sounds the exact same and is not creating an acapella as I was lead to believe.
I have a karaoke version if the song and the original and I aligned them up perfectly and tried several times but it never works.
Is my invert key broken? Am I doing it wrong? Is the website wrong?
Please help as I spent 3 hours + trying to I get it to work but it need did.


:frowning: That’s NEVER going going to work…

If the Karaoke track is the exact same recording without the vocals mixed-in, subtraction would work perfectly. But it’s NOT the same recording. Usually a Karaoke track isn’t even made by the same artist.

If you have a typical Karaoke track that’s different from the original recording, you’ll find that subtraction sounds exactly like addition.* In fact if you record yourself singing or talking, and then you record the same thing again, you voice will be “doubled”. And again, you won’t hear any difference between addition or subtraction. At the sample level (i.e. 44,100 samples per second) it’s all uncorrelated “random” data… Sometimes you are subtracting a negative from a positive, sometimes you are subtracting a positive from a positive, etc., and it’s just so “random” that there’s no sound-difference between addition and subtraction. (Of course, the mathematical values are different.)

Another example would be mixing of a singer and guitar. As you’ve found-out, inverting a single track does nothing to the sound. And if you invert the vocal before mixing with the guitar (i.e. if you subtract the vocal from the guitar) there will be no difference in the mix… The guitar and vocal will sound like they were mixed normally.

Now for example, if have two songs ripped from a CD (they could be two completely unrelated songs) and you mix them together with Audacity, it’s possible to subtract-out one song completely because you have the original byte-for-byte data to subtract. (You have to keep your levels down to prevent clipping during mixing, and you can’t change the levels, etc., but it can work perfectly if you do it right.)


  • Normal mixing is done by summation (addition). Analog mixers are built-around summing amplifiers, and if there are two sounds in the same room the sound waves are summed acoustically.

I disagree (but only a bit :wink:)
The theory is fine, and it will work, but ONLY if the karaoke version is EXACTLY the same as the full version except for the vocal. That is, if both the full version and the karaoke versions were made from the same master recordings, and the mix was exactly the same for all instruments except the vocal, and the audio has not been modified by MP3 encoding, and both versions run at exactly the same speed, and you can align them exactly, and you can match the levels correctly… OK, I’ve changed my mind - it’ll never work :frowning:

Just for interest - here’s an easy example of “cancellation” that does work (totally useless, but proves the theory).

  1. Open Audacity
  2. From the Generate menu, generate a “tone”.
  3. With the track selected, press Ctrl+D to duplicate the track.
  4. With only the second track selected, apply “Effect > Invert”.
  5. Use the track “solo” buttons to play one track, then the other, then both tracks together.

The two tracks are IDENTICAL, except that one is “inverted” - that is, where one waveform goes up, the other goes down (zoom in really close to see this)
When both tracks are played together, they exactly cancel each other out.

You can at least isolate the center in the original.
It holds mainly the lead vocals, drums and some spurious instruments.
In your case, the lead is about 8 dB louder than without isolation.
However, the reverb is removed too and the autotune effect is very audible.
Finding a good voice and record it yourself would certainly be best.