Development of a Two-State Procedure for the Automatic Recognition of Dysfluencies in the Speech of Children Who Stutter, II: ANN Recognition of Repetitions and Prolongations with Supplied Word Segment Markers

The development of an automated method for the assessment of stuttered dysfluencies is described focusing on the second (categorization) stage. The second stage was designed to categorize word level speech as fluent, part or whole word repetitions, or prolongations. Advantages of speech segmentation...

Full description

Saved in:
Bibliographic Details
Published in:Journal of speech, language, and hearing research language, and hearing research, 1997-10, Vol.40 (5), p.1085-1096
Main Authors: Howell, Peter, Sackin, Stevie, Glenn, Kazan
Format: Article
Language:English
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The development of an automated method for the assessment of stuttered dysfluencies is described focusing on the second (categorization) stage. The second stage was designed to categorize word level speech as fluent, part or whole word repetitions, or prolongations. Advantages of speech segmentation prior to processing by an artificial neural network are related. Oscillographic & spectrographic characteristics that define repetitions & prolongations & that distinguish them from fluent words are examined. The ability of artificial neural networks to learn to recognize repetitions & prolongations when they are embedded in fluent word sequences was tested using the characteristics of duration, fragmentation, & spectral similarity. Each artificial neural network was trained using S. E. Fahlman & C. Lebiere's (1990) Cascade Correlation procedure & the human judgment data reported in Howell, et al (1997). The best 10 networks had an overall accuracy ranging from 90.9% to 92%. No single parameter was satisfactory for categorization of repetitions & prolongations. 3 Tables, 5 Figures, 22 References. D. Taylor
ISSN:1092-4388