[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

CASA problems and solutions

Dear List,
   You may have noticed that contributions on CASA have dropped to near
zero after the enthusiasm of the early 1990s. A few people have suggested
that it is time for some serious soul searching on ways to get back on
track.  To address this issue, I offer an essay on problems and solutions
in computational auditory scene analysis at my Web site:
  This essay is intended to be the first section of a longer paper

   In terms of practical results, we must conclude that current CASA
research has been unproductive. I suggest that there has been no systematic
analysis of the basic requirements for how the auditory system has evolved
in the natural environment.  Because of this failure the traditional
research paradigm has ignored fundamental requirements for successfully
operating a _computational_ system in an acoustic scene of many sound sources.
   After analyzing the requirements for separating sounds, I describe a
method for high-resolution temporal analysis.  The system uses a
time-domain method I call interstitial waveform sampling.  The objective is
to obtain extremely fine time resolution while retaining information that
allows intermixed sound sources to be separated into streams from which
meaningful content may be extracted. A few preliminary experiments are
shown here to illustrate the principles of interstitial processing:
     (1) obtaining real-time direction of arrival of the sounds of moving
sources in a noisy, outdoor acoustic environment, and
     (2) using direction of arrival to separate acoustic sources in a
reverberant office environment.

   Best regards,

  John Bates
Time/Space Systems
Pleasantville, New York