[CLAM] clam descriptors computation
xamat at iua.upf.es
Mon May 30 13:46:46 PDT 2005
All of these problems are solved and the descriptor infrastructure is much
better in the upcoming release. We were planning on doing it this week but
we have been very busy preparing the AES convention in Barcelona.
To learn a bit about the descriptor infrastructure you can read this
And this sample application:
And if you can't wait you might try using the tarball that we did for the
LAC, available in clam's main page.
> I have some concerns about descriptors computation in CLAM-0.7.0.
> First of all, segment audio descriptors are not computed when
> SegmentDescriptors::Compute() is invoked. This is because the following
> code snippet is missing from the ConcreteCompute() method:
> Secondly, some of the spectral descriptors are not computed as one may
> expect. I'm thinking mainly about MagnitudeSkewness/Kurtosis, but it may
> apply to other descriptors derived from central moments of the spectrum. I
> discovered and reported this some time ago, maybe a year or so, after
> having inspected the Stats. If I remember right, the skewness/kurtosis is
> computed on spectral magnitudes _only_, and not magnitudes weighted by
> frequency index.
> David: I know your are re-working the descriptors computation scheme.
> Maybe you have planned for these updates? I'm happy to discuss more with
> you outside the list.
> CLAM mailing list
> CLAM at iua.upf.es
Music Technology Group
Universitat Pompeu Fabra
More information about the clam-users