Ableton Forum

are there any good algorithms(ex. matlab) or apps that can separate vocals from a song? i understand this is hard to do, but i think post processing should give much better results than only a filter.

håkan bråkan wrote:are there any good algorithms(ex. matlab) or apps that can separate vocals from a song? i understand this is hard to do, but i think post processing should give much better results than only a filter.

Ahhhh - so you want to remove the egg from a cake which is already baked?
; )

Don't bother - this will sound cruddy at best !

i'm working on remixes for a documentary and i do not have the original tapes, and my intention is also to do some experimental stuff on the material but i would like to have some voices to play with. the original mix is quite thin also, i.e. guitar + vocals, so i think it might work. (if i mask with other stuff in the critical frequencies)

if you have exactly the same song but without vocals you could subtract them [ invert one song ] and get out the vocal.

if the vocal is in the center and the rest of the sounds are not

http://www.ableton.com/forum/viewtopic. ... 2&start=33
that link takes you into the middle of the thread to the crucial point and download, discussion of method follows it.
There is a plugin that extracts the center channel

it will sound terrible, but you can use the general principle to make something a little better (than the demo plugin)

Most vocals have stereo effects on them (e.g. stereo reverb).

Also kick drums and basslines are generally in the centre of the mix also so you'll need HiPass filter the result (some plugins do this as part of the process).

You can pull out the egg out of a cake but you'll get a bit of the flour along with it.

håkan bråkan wrote:are there any good algorithms(ex. matlab) or apps that can separate vocals from a song? i understand this is hard to do, but i think post processing should give much better results than only a filter.

It's always a half-measure trying to remove/isolate vox from a mix and the results will typically sound shit even with £5k's worth of plugs.

But you could try this freebie: Elevayta Extra Boy

http://www.elevayta.com/

HTH

voicetrapp is quite good .

http://www.cloneensemble.com/

Sometimes for issues of time it's easier to just have someone sing it.

håkan bråkan wrote:are there any good algorithms(ex. matlab) or apps that can separate vocals from a song? i understand this is hard to do, but i think post processing should give much better results than only a filter.

No. It's not hard, it's impossible. The only algorithms possible would be based on quantum mechanics. The only current algorithm that works like a charm is: a+b-(o-3 cos)hire-a-vocalist(lmnop-sine<?>^/pi.

I've tried some filters and the extra boy but with bad results as i expected, so maybe there is no other way then to find those hidden tapes ;(

Try audio scanner....think it is by DTG?.....I have gotten better results with that than with extra boy or voice trap......but its beta and has an annoying loud beep every 20sec maybe.....so it depends how long an accapella u tryin to rip......the tools in adobe audition supposedly work relatively well but i dont know whats in their algo outside of the traditional center channel extraction and frequency filter......the added benefit is their spectral tools however, which should help u clean up what the algo missed.....i have actually found it faster and easier sometime to just resing it as close as possible, and then mangle it with FX using formant shifting to change sex etc.......good luck

this is what i used to love about the dancehall and hip hop scenes when i was coming up.....no cds or mp3s....all the hit records u bought came on 45s and 12"......and they almost always came with the version(instrumental) and accapella......u didnt even need a sampler to remix a tune or record a cover.......reminisce