[Clam-devel] Re: GSoC: Enhancing chord detection

[David García Garzón]

Xavier, i just asked Roman to do a quick spike using the implementation he has 
(matlab) to see whether is worth or not to do a new implementation of the 
chord extraction by using such algorithm. Matlab just for the spike.

I fully agree with you that having MP in clam would be nice but that having 
another algorithm not that fast, not realtime and with nearly the same 
results is not worth a GSoC project. So our plan is to quickly explore the MP 
algorithm and if the result are not that precise then go on with the original 
plan.

The spike should have ended today but talking with Roman we have seen several 
things that have not been taken into account for the spike.

In summary, we are on spike and we'll work on MP just if we really get better 
results, at least significantly on accuracy.

David.



On Monday 28 May 2007 19:04:38 Xavier Amatriain wrote:
> Roman, this all sounds great and very promising! I will try to get you
> in touch
> with a PhD student here in Santa Barbara that is doing his PhD on MP
> applications
> for Music. I agree that MP is a cool line of research.
>
> If you haven't done so, I recommend the following papers for alternative
> approaches to Chord Detection:
>
> * A System for the Automatic Segmentation and Classification of Chord
> Sequences by Joshua Morman; Lawrence Rabiner
> * Automatic Chord Recognition from Audio Using a Supervised HMM by Kyogu
> Lee; Malcolm Slaney
>
> They were both presented in a workshop I chaired last year and represent
> state of the art in chord recognition (and are
> both available if you google them).
>
> That said, you must keep in mind that GSoC is a program that is supposed to
> fund development more than research (although it is great when both
> interests
> actually meet!). What I mean by this is that future decisions on how to
> proceed
> should take very much into account how close you are to "implementing"
> something,
> and even more how close you are to making something that can be
> "publicly available".
>
> Doing these experiments in Matlab is great and they should continue but
> I'd suggest
> that you start coding in CLAM in parallel, even if it is just to improve
> minor issues
> related to the current implementation.
>
> Roman Goj wrote:
> > Hi again, sorry for a double post, but I have a correction:
> >
> > On 5/28/07, Roman Goj <roman.goj at gmail.com> wrote:
> >> Ad 2. Well this was a failure :( Admittedly I did not spend as much time
> >> on this - I got too excited with the results from the first try (forgot
> >> to eat my dinner today... this is going to be an exciting summer ;) ).
> >> For this experiment I took 20s out of the example CLAM Annotator song -
> >> Debaser-CoffeeSmell.mp3 and tried decomposing it using mono-frequency
> >> gabor atoms and harmonic gabor atoms (with frequencies f0, 2*f0,
> >> 3*f0)... and mostly what I see in the results are notes right next to
> >> the notes that should be there (the right ones absent) - like in short
> >> time Fourier transform with too short a window - one sees the energy
> >> near one's notes, without being able to perfectly pinpoint the right
> >> ones... Well perhaps this is a bit inconclusive, since I didn't have
> >> enough time to let the scripts work longer on the signal, perhaps over
> >> the night better results will be born (all excited about tomorrow
> >> morning - I'll be dreaming in gabors tonight ;) ). But for now - this
> >> test is a failure :(
> >
> > Correction - I must have been blind and deaf and extremely tired
> > yesterday (didn't seem to be like that yesterday though...)
> > ...because, checking the results of the second try today morning I
> > find there's an almost perfect correlation between my results and the
> > song...
> >
> > A longer - 3 hour trial of harmonic gabor atoms MP in 100 iterations
> > decomposing the first 20s of  Debaser-CoffeeSmell.mp3 resulted in:
> > 97 atoms perfectly fitting the most striking notes in the signal (as
> > judged by both my guitar+ear and CLAM's Annotator :) )
> > 3 atoms showing notes neighbouring the real ones, I think these ones
> > are wrong (unless here my ears can't here them - Annotator shows some
> > faint signs of the notes presence, but I suppose that's just noise in
> > both the algorithms?)
> >
> > I could maybe post the plot when I manage to setup my wiki account...?
> >
> > :) (perhaps my account creation problems have something to do with the
> >
> > new servers?)
> >
> > Also - I have some ideas that should make the algorithm much, much
> > faster and that still in Matlab, so it should work faster when
> > compiled... so with this and the corrected results I dare declare both
> > my trials succesful! :)
> >
> > I'm very sorry for the first misleading post - I didn't check the
> > first results thoroughly enough yesterday :( but I guess it's better
> > this way around than if I first said everything worked great...? :)
> >
> > Roman
> >
> > _______________________________________________
> > Clam-devel mailing list
> > Clam-devel at llistes.projectes.lafarga.org
> > https://llistes.projectes.lafarga.org/cgi-bin/mailman/listinfo/clam-devel
>
> _______________________________________________
> Clam-devel mailing list
> Clam-devel at llistes.projectes.lafarga.org
> https://llistes.projectes.lafarga.org/cgi-bin/mailman/listinfo/clam-devel