Extracting Vocals
This article was originally posted by Daz of DazAutomatic April 9, 2003
Extracting Vocals/Acapellas From CD Singles
This article is based on a short tutorial by Dsico on extracting vocals from songs. The key to success with this method is having a CD single that features an instrumental version that is exactly the same length as the main edit. I have used this method on a few singles successfully, and this following article talks through the process, using Sonic Foundry's Sound Forge 6.0.
Please don't email me telling me that acapella is actually two words. I don't care, it's not anymore.
There are several ways to get your hands on an acapella version of a song. Obviously if a single is released with an acapella (sometimes on a vinyl release you'll see an acapella) you can just get a hold of that, or check out McSleazy's forum for friendly bootleggers giving out acapellas. A lot of people find SoulSeek useful for finding acapellas.
So you can't find that copy of the Gospel Choir sings the hits of Take That. You've searched the forums and the internet and you can't find those vocals anywhere. Looks like your going to have to create that acapella yourself. As a rule it's impossible, and you're stuck. But you might be lucky if you find the single happens to have a radio edit and an instrumental of the same length. The theory behind the method relies on the fact that the only difference between the radio edit and the instrumental is that the engineer muted the vocal channels in the mix when mixing down the track. So if you've got the required media and a copy of Sound Forge (of course you can use another editor, but Sound Forge is the best sample editor ever created), we're ready to begin...
Step 1) Prepare the two tracks
You'll need to rip the radio edit and the instrumental from a CD. I've tried sampling from vinyl with little success, and i've been told that working from MP3s doesn't work either. It's believed that this is due to sampling and/or compression adding articles to the waveforms.
When you've got them into Sound Forge you'll need to align the start of the two files. First switch so that the time display in the sample edit windows are both showing sample no.s and not minutes and seconds or anything else.
The image above shows where Sound Forge displays the sample numbers, the first is the start of the selection range, the second shows the sample number of the end of the sample range and the third number shows the number of samples within the range of the selection.
In order to line up the two tracks exactly you should zoom in at the start of the track an look for a glitch or recognisable peak in the waveform that features in both tracks. The next picture shows an example...
You will need to zoom in until you can be certain that the cursor in each case is exactly over the peak. It is probably a good idea to insert a marker here for future reference (just press M). Read of the cursor positions from the bottom right corner of the window. If the two numbers read the same then you are lucky, and the tracks are lined up perfectly. It is probably the case that one of the sample number readings is larger than the other. In this case you should calculate the difference between the sample numbers and remove that many samples from the beginning of the track. An easy way to do this is to insert a marker, and then double click to set it's properties, as shown by the next example...
In the field labeled start you should insert the difference that you calculated that will align the two markers for the peak. Once the marker is in the correct position you can delete all the data that precedes it. Now check that the two markers you placed to mark the peaks in the waveform are exactly lined up.
Step 2) Phase Reversal
If you've got this far congratulations, the hard bit is over. The next step is to invert the instrumental and mix it with the vocal version so that the instrumental parts all cancel out and leave just the Vocal. Sound Forge provides a neat method of achieving this.
Select the entire instrumental track and copy this to the clipboard. Then select the entire vocal radio edit version. You should then select the Paste/Mix tool which looks like this...
Notice that the Invert data box should be ticked. This performs the phase inversion required for the cancellation to take place. Once you selected this you can click OK, then hopefully you should have successfully created an acapella.
I get a strange phasing noise…There's many reasons why this happens. When the samples aren't quite lined up you get this effect. If they at slightly different speeds then this can be the result. Check the alignment of the two tracks. Try aligning them against a mark later in the track. I found that when working on J.Lo's "Jenny from the Block" I found that aligning them from the start didn't work, but when aligning them from the start of the rap after the intro noises it seemed to work fine.
If the reason is because of a slightly different bpm then there's not much you can do.
It works for the first few seconds but then the instrumental fades back in…Probably because your tracks are at slightly different speeds, or there's some variation in the edit. You could try cutting the track into smaller samples and then performing the process on smaller pairs of samples. Or try loading the two sources into acid as "one shot" samples, you can then cut and line them up quite easily.
Good Luck!