[CLAM] ClamExtractorExample LowFreqEnergyRelation

David García Garzón dgarcia at iua.upf.edu
Mon Feb 25 06:16:41 PST 2008


On Monday 25 February 2008 12:25:24 vincenzo.dimattia wrote:
> Dear all,I am Vincenzo Dimattia, I am developing my final project, an
> automatic segmentation system,and I am using the programm
> ClamExtractorExample for the extraction of the audio descriptors . I would
> like to know how is calculated the LowFreqEnergyRelation.This should be
> higher for low frequency signal and lower for high frequncy signal. Is it
> right? Thanks
> Vincenzo Dimattia

Hi, Vincenzo.

There was an error on doxygen description. LowFreqEnergyRelation had the 
description of MaxMagFreq. I just fixed it on the svn. The description for 
LowFreqEnergyRelation is now this one:

"Ratio between the energy over 0-100 Hz band and the whole spectrum energy.
To avoid singularities while keeping descriptor continuity, when the whole 
spectrum energy drops bellow $10^{-4}$, such value is considered as whole 
spectrum energy."

So you are right with your interpretation. But keep in mind that 0-100 Hz is a 
very narrow band and it is used to spot bass sounds. If you are interested in 
locating high frequencies, HighEnergyContent is more appropiated. It is 
defined as:

"Sum of the squared spectrum magnitude multiplied by the wave number of the 
bin. It is pretty similar to the derivative of the energy, or a high pass 
filter, which gives higher values for high frequency content."

We would be glad if you report back to the mailinglist your progress with your 
final project :-) And, please, consider also integration of your work into 
CLAM code base. Integration into a big existing project is often appreciated 
by evaluators. Just drop us a line if you want to.

David.






More information about the clam-users mailing list