Sound source separation plugin

Special Interest Groups Audio Processing
1

I’m not sure it’s possible to be in Audacity, I’m not sure it already does exist as addon plugin, but it would be nice to have feature to separate various sounds to individual tracks… Like other sound separators…

2

Like other sound separators…

Such as?

As a rule you can’t split a mixed performance up into individual instruments and voices. One of the engineers at work had a program that would do a remarkably good job, but you would never mistake the separated track for a natural track or instrument. It sounded a little like a ratty cellphone call.

He was working from a very high quality Music CD. It’s much worse if you start out with a highly compressed internet MP3. Not only does the software have to split apart the voices, but also the music from the compression damage.

Koz

3

I can’t remember software names for sound source separation process, although many of them are not free…

I was only asking is it possible to port that function to audacity. It doesn’t have to be so much quality as the commercial applications, but some of other sound editors, especialy the commercial, have that feature in basic installation.

4

Do you mean http://wiki.audacityteam.org/wiki/Proposal_Source_Separation ?

Or Audio Diff/Pattern Matching http://wiki.audacityteam.org/wiki/Proposal_Audio_Diff ?

Describe what these features do, if you can.


Gale

5

Yeah, something like that…

I’ll explain it on an example:
I got recording of an animal sound. There’s a background noise (some other animals, forrest noise, etc.), but I want the desired animal sound only. I would use the sound source separation process to make several tracks from one sample and keep only the one, with that animal…

Another example: Recorded few people talking in a conversation and want to extract only two people talking within them. The same process should give me separated tracks of each person in the disscussion. I’ll find people talking within them and keep only that tracks…

I hope this would explain what I’m looking for…

6

I’ve seen some experimental software that can do a reasonably good job at separating two voices from a clear high quality stereo recording of two people that were talking from different places in a quiet room. CSI and James Bond show this technique working much better than I have seen it work in real life.