G'day to all,

Anyone knows how to remove (most of the) music from a song and keep only the vocals?
Removing vocals is easy, this could be harder, right? Thanks .........

Ummmm...
Why???

(Welcome to the BR, by the way)

A mate of mine was playing around with this, to get some vocals of Shaggy for a radio programme project - as you say, it's easy getting rid of the vocals, but leaving the vocals is quite a bit harder.

I think he ended up using the vocal remove in Adobe Audition, inverted it, and mixed it with the original. I think he managed it, but his hairline is now way back!!

Hi,

The other option is to use the function in audition has the option to take a sample segment of audio, and then remove that sample from what ever you apply it to. (I understand its designed so you can find a section with just noise on it, and then apply it across the whole track). AFAICR its in the noise remove section. So if you rmanaged to get just the backing and then sampled that it might be worth a try.

As with all these things though it's less than perfect, I know I've probably shaved a few years off my life trying to do similar things for radio imaging! Good Luck!

Lewis

What you're asking to do is roughly akin to trying to unbake a cake and recover the eggs. It's never going to work very well.

As has already been mentioned, later versions of Adobe Audition (AA1.5 and AA2.0) have a feature called "centre channel extractor" which at least makes an attempt at doing this. How well it works depends very much on your source material though.

All methods of removing voice work basically the same way: they take advantage of the fact that the voice is USUALLY panned to the centre of the audio image with the backing tracks spread across the sound stage. If this IS the case then inverting the waveform allows the voice to be cancelled, in theory without affecting other parts of the track. The "centre channel extractor" takes this a stage farther and basically uses the technique jayselway mentions but does it automatically. The Audition system also gives you lots of user controls to tweak this process for optimum results.

However, you can probably see that this method is usually going to be far from perfect. First, other instruments, particularly drums, are often panned centrally and therefore affected by the process. Second, reverbs and other effects added to the voice are often panned left and right even if the main voice is central. This leaves the ghostly reverb present even if the central voice is gone.

By all means try Audition and see if it does what you want on the track you need...it's probably the best option there is. However, do be aware that even with Audition the results will be variable.

Also within Audition, Big_L suggests using the Noise Reduction function. This won't work for what the OP wants at all...it's designed to remove general background noise from tracks but is only effective if the "noise" is fairly constant. Clearly, a music track is far from this! However, there is one other feature that might be worth playing with. Again, in AA1.5/2.0 there's a feature called "Frequency Space Editing". This allows you to go to spectral view and actually draw around a range of frequency/time, then manipulate your selection (including simple deletion). By itself this won't do what you want since the frequeny ranges of voice and music overlap, but it MAY allow you to add some refinement to the intial centre channel extraction mentioned above.

Bob

It might also be worth finding out whether there was an acapella version of the song you want to use. Sometimes you'll find one of the bonus tracks on a CD single is an acappella version of the main track while other times record companies will release acapella versions to DJ's so that they can mix the track in with something else.

Cheers

James.

I always say it's getting the milk out of a cup of tea!

they'll steal that, and the sugar around here, so maybe we should ask around.

Ebay usually has a good selection of acapellas or people with access to them !!

I used Fleximusic Wave Editor using Band Pass Stop Filter in that you can either do
blocks or keeps the frequency range. But you have to find by trial and error in which frequency bands to pass/stop.

Although I've tried a zillion ways to remove the centre info over the years with (as everybody else who's tried to say - very indifferent and unreliable results) little success, I'd never thought about trying to then matrix out the vocal - so I have had a go.

what I did was take a small snip from an Eva Cassidy song - complete with hiss, because it's typical - and only has a voice and guitar - so far, far the simplest material to process.

First treatment was to remove the centre channel, karaoke style, and then take the two and matrix out just the original vocal

here are the versions

original
original with vocal removed
matrixed out leaving vocal

The problems are quite obvious. The reverb wrecks the vocal removal - the hiss isn't a major problem.

The end result is pretty poor - certainly not usable. As everyone has said - a totally unpredictable and unreliable process.

For info, the vocal removed in audition, then the matrixing by cubase sx - all from 44.1K wavs from the CD, only converted to mp3 for the web - the artefacts heard are all present in the uncompressed format - horrible!