From: malcolm@interval.com (Malcolm Slaney)
Date: Mon, 29 May 1995 11:39:46 -0700
Subject: Perceptually Based Audio Quality Measures
Message-Id: <abefc3f0050210042e9f@[199.170.108.19]>

This week at the CCRMA Hearing Seminar, Bill Putnam continues last week's
discussion with a review of perceptual audio measures.  Just how should we
measure audio quality?  Nobody can afford to put lots of audio experts
(i.e. golden ears) in front of all their audio work.  Can we build computer
algorithms that will simulate the golder ears?  Are the algorithms accurate
enough so that engineers can use them to build better coders?  What's
missing? How well do they work?  Bill will review the current state of the
art and lead the discussion about interesting direction to go next.

Bill Putnam has been a student at CCRMA for the last couple of years and a
frequent attendee of the seminars.  Come here his ideas about perceptually
based audio quality measures.  He's bringing audio examples to illustrate
the problems and ideas!

        Who:    William Putnam (CCRMA)
        What:   Perceptually Based Audio Quality Measures
        When:   Thursday June 1 at 11AM
        Where:  CCRMA Library (Top Floor of the Knoll at Stanford)

Come hear the perceptual coders and their measures at CCRMA!

-- Malcolm



Engineering Applications of Perceptually Based Audio Quality Measures
Bill Putnam (Stanford CCRMA)

Auditory based perceptual issues are becoming increasingly important
in the context of audio engineering.  Speech and audio compression
are perhaps the most notable examples, but recent literature in other
areas such as filter design and loudpeaker arrays have indicated an
increased attention to perceptually based issues.

Traditional engineering tools such as signal to noise ratio are clearly
insufficient in predicting subjective qualilty.  As a result, subjective
listening tests have traditionally been necessary to quantify results.
Due to the high cost of conducting such tests, it became evident that
objective measuring techniques which could predict the outcome of these
tests would be useful to researchers.  Over approximately the last 10 years,
several researchers have addressed this issue[1-4].  These techniques
typically use auditory models to either predict a threshold of audible
distortion[3], or establish a metric to measure the difference between a
reference signal and a test signal[2,4].  I will attempt to present an
overview of some of these techniques.  Additionally, I hope to entertain
a discussion concerning the possible applications of these methods to
engineering problems, not limited to compression.

[1] M.R. Schroeder, B.S. Atal, J.L. Hall, "Optimizing Digital Speech
Coders by Exploiting the Masking Properties of the Human Ear," J.A.S.A.,
Vol. 66, pp. 1647-1652, Dec. 1979.

[2] Matti Karjalainen, "Sound Quality Measurements of Audio Systems
Based On Models Of Auditory Perception," Proc. ICASSP 1984, pp. 211-215.

[3] Karlheinz Brandenburg, Thomas Sporer, " NMR and Masking Flag:
Evaluation of Quality Using Perceptual Criteria,"
Proc. of the AES 11th Intl. Conf, 1192, pp.169-176.

[4] John G. Beerends, Jan A. Stemerdink, "A Perceptual Audio Quality
Measure Based on a Psychoacoustic Sound Representation,"
J. Audio Eng Society, Vol. 40, No. 12, Dec. 1992.