|
Re: Below, while I think a frequency tiling that is somewhat
like a wavelet is quite appropriate, in fact something more like a tiling of .7
Bark at low frequencies and 1 ERB at higher frequencies, with attention paid to
the slope of the filter skirts, would be ideal, I don’t think a wavelet
is idea for auditory analysis, since the ear consists of a set of heavily
overlapped, far from 1:1 and onto “bands”. I know from
working on loudness models that you must have a filter at each 1/3 ERB (at
higher frequencies, let’s stick to something like .7 bark at low, please)
with the appropriate response CENTERED on that frequency. Using minimum-phase
seems ok for this. But, with wavelet transforms, you’re going to have 1:1
performance, and spacing accordingly, with critical sampling properties. This
is not going to handle edges of bands very well in my experience. In particular, I’m objecting to the 1:1 and onto
properties of the wavelet, they do not match how the ear works. Of
course, if you must do exact reconstruction, that’s a different issue. __________________________ James
D. Johnston (jj@xxxxxxx) CHIEF SCIENTIST - DTS, Inc. Those interested in the mathematical basis of phase
perception might like to look at a paper by Martin Reimann that appeared in
JASA a few years ago. After demonstrating that the cochlea preforms a wavelet transform
rather than a windowed Fourier transform, he goes on to describe how phase
operates in the wavelet representation of auditory processing.
Notice:
This message and any included attachments are intended only for the use of the addressee, and may contain information that is privileged or confidential. If you are not the intended recipient, you are hereby notified that any dissemination, distribution or copying of this communication is strictly prohibited. If you have received this communication in error, please destroy the original message and any copies or printouts hereof. |