# [music-dsp] Inverse karaoke

Pavel Savygin p.savygin at virtualphotonics.com
Sat Mar 3 04:14:31 EST 2001

```Hello Pasta,

Thursday, March 01, 2001, 6:28:05 PM, you wrote:

PB> I would like to ask if anybody knows of any device, system, algorithm, or
PB> whatever, that does exactly the inverse of a karaoke function: i.e., instead
PB> of supressing the vocals from the music, I would like to suppress the rest
PB> and leave the vocals.
Algorithm is very simple in its basic definition. It bases on
assumption that voice is recorded in the middle of two cahnnels. So if
we want to extract music - we are just to substract one channel from
another. But this will wotk correctly only if every instrument is
mixed only in one channel. And the voice is in the middle. That means
that for voice we have equal values in both channels. For example we
have music with drums(d), guitar(g) and voice (v)
SO let us assume:
left channel=d+v/2;
right channel=g+v/2;
left channel-right channel=d-g;
For guitar, fhase will be inverse, but it means nothing for our ears.
But this equations does not work if phase of voice is moving diffrently for
every channel. This alwats occurs for bad encoded mp3.
And for extracting voice I see no way except finding out FFT for
(d-g). and substracting absolete values of FFT(d-g) from
FFT(left+right). Not good idea but it must work.

PB> I don't need this to be really accurate; I'd be happy with something that
PB> would allow me to attenuate everything but leave the vocals. I need this to
PB> perform some psychoacoustics' experiments involving attention to singing
PB> style.

PB> Any ideas, pointers to references, etc will be most welcome.
PB> Thank you advance.

--
Best regards,
Pavel                            mailto:p.savygin at virtualphotonics.com

dupswapdrop -- the music-dsp mailing list and website: subscription info,
FAQ, source code archive, list archive, book reviews, dsp links
http://shoko.calarts.edu/musicdsp/

```