[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [AUDITORY] review of harmonic+noise segmentation software



Matt,

A good historical overview of sinusoidal modeling which includes sines+noise and a few extensions is here: https://ccrma.stanford.edu/~jos/pdf/SMS.pdf
SPEAR is a popular choice for sinusoidal modeling (without noise): http://www.klingbeil.com/spear/
Professor DAn Ellis has some Matlab code for doing sines+noise modeling: http://www.ee.columbia.edu/ln/rosa/matlab/sinemodel/
I don't know of any software packages which perform, for example, sines+noise+transient modeling but I imagine they must be out there.

A related technique which has seen a lot of focus is harmonic/percussive separation or transient/steady state separation.  There have been many techniques proposed; my favorite is FItzgerald's method because of its simplicity and effectiveness: http://arrow.dit.ie/cgi/viewcontent.cgi?article=1078&context=argcon
There is an implementation available in librosa: https://github.com/bmcfee/librosa/blob/master/librosa/hpss.py

Hope this is helpful.

-Colin


On Mon, Sep 30, 2013 at 7:48 AM, Matt Flax <flatmax@xxxxxxxxxxx> wrote:
Hi there,

Back about 10 or 15 years I remember that some people on this list were researching the segmentation of sound into tonal components and noise components.

I was wondering whether anyone on the list has used any free software libraries which come from this research to do this and if so whether you could let the list know ?

I am intersted in software which can do more then monophonic segmentation ... if possible.

thanks
Matt