Classification of Speech Sound Errors


Outline of Research

An automatic classification tool for speech sound errors was 
developed. Students speak short words, representing vowel sounds 
in the English language, into a microphone. The tool, in form of a 
computer program, detects and classifies possible vowel substitutions 
automatically in real-time. The methodologies for sound acquisition, 
signal pre-processing and classification are outlined. A satisfactory 
correct recognition rate of 80% to 90% was obtained, with a
typical classification time of 1 second.

	Participating in the project were Dr. Bernd J. Kurz (professor), 
David Gregg (formerly MCS student) and Cynthia Howroyd (The Speech 

	This research was carried out in collaboration with The Speech 
Institute Ltd., Fredericton, N.B. The classification tool is to be 
licensed to The Speech Institute.

Project Status: Started:   March 1995
                Completed: January 1996

Relevant Publications

Kurz, Gregg, Howroyd: Real-Time Automatic Classification of
  Speech Sound Errors, Proc. Int'l. Conference Natural
  Language Process. and Industrial. Applic., NLP+IA’96,
  GRETAL, Univ. of Moncton, Moncton, NB, Canada, 4-6 June
  1996, Vol.1, pp.141-147.

David Gregg, The Automatic Classification of Short Speech Sound 
  Deficiencies, MCS thesis, Fac. of Computer Science, Univ. of New 
  Brunswick, January 1996.
  (Thesis is classified and on embargo until February 1998).


Back to: Kurz's Research Page | Kurz's Home Page

Last revised: 6 June 1997 by Bernd Kurz bjkurz@unb.ca