[sdiy] Waveform Classification System (timbre index???)

[Chris Stecker]

oops...sent this out to AH on accident.  

> > Has anyone seen any research on classifying waveforms musically or
> > sorting them by some kind of timbre index?? (Or has anyone seen the
> > phrase "timbre index" before?)

I'd say David Wessel (now at CNMAT, the center for new music and audio 
technologies) is your man for timbral classification and control.  Wessel was 
working to apply results from multi-dimensional-scaling studies of timbre 
perception (Grey 1975) while at Michigan State in the early 1970's, and 
continued in that line of research at IRCAM in Paris and later as the 
director of CNMAT.  Much of his work has been aimed at developing 
low-dimensional, musically-controllable, representations for many-parameter 
synthesis systems, i.e. additive.  It's work aimed at computer-based 
synthesis, and generally reflects the desire to recreate and modify acoustic 
timbres, but most of the work should apply (a big part was simply the 
development of methods for obtaining similarity matrices and transforming 
them to musical control space).    Namedrop factor:  Wessel was one of the 
signers of my dissertation, as well as a friend and collaborator of Don 
Buchla's.

Here's a good overview of some early parts of Wessel's "Timbre Space" story, 
available online from IRCAM:

Wessel, D.  1979.  Timbre Space as a Musical Control Structure.
Computer Music Journal 3(2).  
http://catalogue.ircam.fr/articles/textes/Wessel78a/

Later work focussed on correlating the principal components of the recovered 
timbre-space with acoustical and/or physical properties of the stimuli.  
Highly-determinative acoustical properties included envelope characteristics, 
"brightness," "temporal flux" (differences in the temporal envelope of 
different harmonics), etc.  One approach would be to simpy correlate your 
waves with these parameters and use the correlations to place timbres in 
space, but this probably won't work well for several reasons I could go into, 
but won't.

The "correct" thing to do is probably to analyze some judgments of timbral 
differences among the set of waveforms of interest, using the method of 
multi-dimensional scaling.  If you're interested in doing this, I'd be happy 
to consult/collaborate.  It will be complicated (or maybe simplified) by the 
fact that we normally use filters to control brightness and EG to shape the 
envelope.   We can control those parameters to acheive a more sensitive 
description of the steady-state aspects of the waves.  Perhaps there's a way 
to use the AH/sdiy list membership as a source of listening subjects (indeed, 
"expert" listeners when it comes to synthetic analog waveforms).  

The outcome of constructing a timbre space for mini-wave waveforms would be 
an organization where some number of CV inputs (1-3?) can be used to search 
through the timbre-space.  A small change in voltage will produce a small 
perceptual change, large changes likewise, and changes on the different 
inputs will tend to produce "perceptually-orthogonal" changes.  This would be 
pretty great, and it would be quite interesting to consider how later 
processing (filtering, modulation, etc) comes to deform the timbre space, and 
how it might be combined with "timbre-index" (sine-indexed wavetable) style 
use of the mini-wave.  Exciting!

-Chris


-- 

G. Christopher Stecker, Ph.D.                                              
cstecker at umich.edu	/    734-764-5167

Central Systems Laboratory, Kresge Hearing Research Institute                 

University of Michigan Medical Center
1301 East Ann Street / Ann Arbor MI 48109-0506