This should be easy, but I’m stumped. Need advice please!

So from your problem I understand that you are trying to obtain the acapella by adding inversions of the tracks you have to one another?

Try this:

Get your ‘overlapping acapellas’ track

duplicate it

Invert the duplicate

Time shift the duplicate by how however much the two acapellas are offset.

Now play the two tracks together and hopefully have an acapella?

Let us know how you get on…

(You can ignore my original posts but I will keep them up as they might be useful for future reference)

***Original Post 2
For it to work though the clean and instrumental signal have to be precisely the same just time shifted by a certain amount of samples (ie individual signal values)
And the method I describe will cease to work once the two acapella signals are no longer identical (ie there is a censored word)

To demonstrate I will use a simple example signal (which is the equivalent of your overlapping acapella):

Imagine there is a ‘combined’ signal made up of two summed ‘original signals’ which are identical but of which one has been time shifted ‘to the right’ by two samples.

Imagine the ‘combined signal’ has these values:

3, 4 , 8, 9 , 9 etc

From this we can see that the first two samples of the combined signal are 3 and 4.

Therefore we know the first two samples of the original signal are also 3 and 4 as the time-shifted original signal has not started.

Now as we know the time shifted signal starts on the third sample (as it has been timeshifted two samples) we can subtract the value of the first sample of the untimeshifted original signal from the third sample of the combined signal to obtain the value of the third sample of the untimeshifted original signal. This process can then be repeated for the rest of the signal.

The main problems in practice would be finding how many samples the signals had been time shifted by, and whether the time shifted signal was indeed identical.

Also there is a problem with calculating the valued once a word has been censored from the acapella but I think this could be overcome, ie by looking at periods of silence, or maybe finding values by trial and error.

So, Are you any good at programming?
***Original post 2 end

Original Post***

From my maths knowledge I think it is impossible to do this and it can be illustrated through simultaneous equations:

In the equations below Signal 1 is the ‘Clean Edit’, Signal 2 is the ‘Dirty Edit’ and Signal 3 is your overlapping acapella.

Music is the Instrumental content of your track and Vocal A and B are the acapella contents of each track.

Music + Vocal A= Signal 1 <<<Clean Edit
Music + Vocal B =Signal 2 <<<Dirty Edit

Vocal A + Vocal B =Signal 3 <<<Overlapping Acapellas

Standard simultaneous equation theory dictates that you can only solve a number of unknown variables equal to the number of equations you have.

In these equations we have three unknowns; Music, Vocal A and Vocal B. We know these are unknown as if we knew them we wouldn’t even be undertaking this task .

And while it seems that there is a third equation, the overlapping acapella equation, that is simply obtained by combining the previous two equations (known as a linear combination of the other equations) so is of no use.

I think this right…more than happy to be corrected though ***Original Post ends