Re: soundscape analysis? Get real!

Subject: Re: soundscape analysis? Get real!
From: Jean-Marc Pelletier (
Date: Fri Dec 17 1999 - 22:07:02 EST

About wavelets. . .
FFT graphical analysis seems to me terribly biased towards certain types of
sounds. That is one can easily distinguish an A440 note played on an oboe
and the same note played by a clarinet with such a graph.

However, the original post was about soundscapes. My ears, and my mind, tell
me that the sound of waves crashing against a fine sand beach and that of
waves rolling against a pebble beach are quite different, as much, if not
more, as an oboe differs from a clarinet. However that difference would
never show up on an FFT graph. Or at least not in a meaningful way.

Natural soundscapes are made up mostly of noise spectra and to analyze those
using the FFT, which is pretty much the only method available to the average
electroacoustician, seems like using a hammer to screw a nail.

The question I was asking myself is: would wavelets do a better job at
analyzing and differentiating such spectra?
If not, what research has been done in that way?

Jean-Marc Pelletier
Nanto, Japan

-----Original Message-----
From: Linda A. Seltzer <lseltzer@phoenix.Princeton.EDU>
To: <>
Date: December 17, 1999 6:57 AM
Subject: Re: soundscape analysis? Get real!

>The Matlab wavelet toolbox can be used to do CWT or
>FWT analysis, so the software is no problem, except that
>when I studied that package I found a bug in which
>some of the arguments to one command were being called
>in the wrong order. I reported it to the Mathworks
>and they verified it, so I hope it's not present
>in current versions. (This happened in 1997). If I
>have time I will see if I still have any notes on this.
>Also, some of the wavelets generated in Matlab are
>normalized and some of them are not. You have
>to study the package a lot before using it.
>The other problem is this: Intuitively we can correlate
>frequency with our perception, but do we really have
>a feeling for a scalogram? The graph of a CWT will
>tell us which translations and scales have high
>correlation with the signal, but beyond that what
>do you know intuitively from it? What is the
>wavelet analogy to a formant? This is why you
>don't see very many wavelet analyses around.
>An FWT filterbank may have interesting analysis/synthesis
>properties, but is there a wavelet synthesis method as
>an analogy to SMS with sine waves? Wavelets might be
>interesting with granular synthesis, but I have never
>had the time to try it. Barry Truax, if you are
>reading this, have you tried wavelets as grains?
>Did it make a difference as compared to other types
>of signals as grains?
>Linda Seltzer

This archive was generated by hypermail 2b27 : Wed Jun 11 2003 - 13:09:17 EDT