sound source separation ( RE: [music-dsp] complicated editors )
joshscholar at yahoo.com
Sun Feb 20 20:42:29 EST 2005
----- Original Message -----
From: "robert" <r_obert at hotmail.com>
To: "'music-dsp'" <music-dsp at ceait.calarts.edu>
Sent: Sunday, February 20, 2005 4:47 PM
Subject: sound source separation ( RE: [music-dsp] complicated editors )
> > -----Original Message-----
> > From: music-dsp-bounces at ceait.calarts.edu [mailto:music-dsp-
> > bounces at ceait.calarts.edu] On Behalf Of Joshua Scholar
> > Sent: Saturday, February 19, 2005 8:56 PM
> > To: music-dsp
> > Subject: Re: [music-dsp] complicated editors
> > The near impossible case is blind separation (where you're trying to
> > completely separate one instrument from another).
> >From my little bit of investigation, I agree.
> > One element of separation I wonder about is the use of inverse comb
> > filters
> > (whatever that's called) so that you could include partials from the
> > target
> > sound even though they overlap the other instruments or notes.
> Nice idea. Do you know of any publicized research into this?
A little while back I read a blurb saying that LMS adaptive filters have
trouble optimizing inputs that are highly coorelated. That got me thinking
about partials and inverse comb filters...
A couple of weeks ago I tried to write a decliper that attempted to use an
all poles filter to bridge the clipped areas. I started with a
generalization on linear prediction that allows gaps in the input... but no
matter what cludgy algorithm I wrote to move poles around the results were
bad. Extrapolating filters are hard!
This got me thinking about that most powerful of optimizing algorithms -
having a human being tweak the controls and examine at the results until he
likes them. If you have a good user interface you can design an algorithm
by doing it by hand for a while and then enshrining your experience in a set
And thinking about all of the algorithms I could make up THAT way got me
interested in writing an graphic sound editor.
I hope I get a chance to finish that editor I started... Practical problems,
like other work or needing money may intervine.
> No virus found in this outgoing message.
> Checked by AVG Anti-Virus.
> Version: 7.0.300 / Virus Database: 266.1.0 - Release Date: 2/18/2005
> dupswapdrop -- the music-dsp mailing list and website:
> subscription info, FAQ, source code archive, list archive, book reviews,
More information about the music-dsp