From: malcolm@interval.com (Malcolm Slaney)
Date: Thu, 29 Sep 1994 22:25:00 -0700
Subject: Onsets at CCRMA
Message-Id: <9409300522.AA08511@interval.interval.com>

***** Note, we're moving the Hearing Seminar back to 11AM.  The undergrad
      psychoacoustics class is meeting from 9 till 11, so we'll have a bit
      less conflict!

Next Thursday, Richard O. Duda will be talking about work he and Tareq
Shahwan have done to build a computational model of onset detectors.
Certainly common onsets are the most important clue for auditory grouping.
If two components of a sound both start at the same time, then we're
probably going to hear them as part of the same auditory object.  This true
for both harmonic and inharmonics sounds.  Onsets are a powerful grouping
mechanism.

All of the sound separation systems that have been described at CCRMA
during the past few weeks have had some notion of common onsets.  But what
does an onset really mean?  The neurophysiologists have their own
definition, is that the right definition for auditory grouping systems?
Dick will describe his work, and present examples.  Come to CCRMA to judge
for yourself.

        Who:    Richard O. Duda (SJSU) with Tareq Shahwan
        When:   11AM, Thursday October 6    <====== Note new time!!!
        What:   Acoustic Onset Detection
        Where:  The CCRMA Library (Top Floor of the Knoll)

We're getting back to basics.  First onset detectors, and hopefully some
neurophysiology later in October.  Bring your questions to CCRMA!

-- Malcolm



----------------------------------------------------------------------
        Adjacent-Channel Inhibition in Acoustic Onset Detection
                    Richard O. Duda and Tareq Shahwan


                             Abstract

Simple acoustic onset detectors respond to sudden changes in energy
levels.  Unfortunately, they respond to frequency modulation as well as
to amplitude modulation, and have high false-alarm rates in the presence
of siren or chirp signals.  We show how adjacent-channel inhibition can
be used to greatly reduce the false-alarm rate.  An LMS procedure is
used to obtain optimal inhibitory eights.  Computer simulations reveal
how the response to chirps varies with the rate of change of
instantaneous frequency.
-------