From: malcolm@interval.com (Malcolm Slaney)
Date: Sun, 10 Dec 1995 19:01:05 -0800
Subject: Localization of Sound Sources
Message-Id: <v02130504acf14e56e793@[199.170.106.94]>

Perhaps one of the hardest unsolved problems in audition is understanding
how humans localize auditory signals.  With just one or two sensors, how
can we possibly know where a source we've never heard before is located?
We all know that timing tells us part of the story, but how do we tell
elevation?  Or even harder, how do we tell elevation in the plane between
the two ears when we hear a monaural signal?

Dick Duda will be addressing this topic at the Hearing Seminar on Thursday.
He has been building models of human localization, and trying to
understand the cues that are important.

        Who:    Richard O. Duda (SJSU)
        What:   Localization of Sound Sources
        When:   Thursday December 14 at 11AM
        Where:  CCRMA Library (Top Floor of the Knoll at Stanford)

This is the last Hearing Seminar of the year.  See you at CCRMA!!!

-- Malcolm


         Combined Monaural and Binaural Localization of Sound Sources

This talk is an based on work that  William Chau and I presented at the
recent Ailomar Conference.  The central idea is to combine monaural
estimates of elevation with binaural estimates of both azimuth and
elevation to exploit the special advantages of each.  The monaural
estimates are derived from the second difference of the amplitude spectrum,
and are useful in the median plane, where binaural elevation estimates are
very poor; however, they are source dependent, and do not provide azimuth
information.  The binaural estimates are derived from the interaural
intensity difference, and are accurate away from the median plane.  For a
white-noise source in an anechoic environment, the average error for the
combination is about 2-degrees for azimuth and 12 degrees for elevation.
Good performance is also obtained for speech, but accuracy degrades as the
source bandwidth is reduced.