Newsgroups: comp.speech
Path: pavo.csi.cam.ac.uk!doc.ic.ac.uk!agate!ames!sun-barr!cs.utexas.edu!torn!spartan.ac.BrockU.CA!bradford
From: bradford@spartan.ac.BrockU.CA (James Bradford)
Subject: Measuring Speech Duration
Message-ID: <1993Jan12.211558.6884@spartan.ac.BrockU.CA>
Organization: Brock University, St. Catharines Ontario
X-Newsreader: Tin 1.1 PL4
Date: Tue, 12 Jan 1993 21:15:58 GMT
Lines: 29

A few months back I posted a request for information on algorithms 
for measuring the duration of spoken utterances.  I received a number 
of very helpful replies.  The paper I found the most useful was:

     Rabiner, L.R. and Sambur, M.R., "An algorithm for 
     determining the endpoints of isolated utterances," The Bell 
     System Technical Journal, 54(2), 1975, pp 297-315.

The paper contains a couple of typos that slowed us down a bit.
If anyone else plans to implement the Rabiner algorithm, I'll
be happy to provide our corrections.  

In the last couple of days I have run 3 subjects with 10 trials 
each (the input text was "The quick brown fox jumped over the 
lazy dog").  This (very) preliminary pilot study suggests our 
subjects have a durational precision of +/- 5% of their mean 
duration.  If the results hold for a statistically valid sample,
I wonder if a small, speaker independent recognizer with a 
carefully chosen vocabulary might be based on duration alone?  

Anyway, my thanks to everyone who replied.  Your help was much 
appreciated.


James Bradford, Director
The Speech Interface Lab,
Brock University
EMail: bradford@spartan.ac.BrockU.ca

