Development of a Two-Stage Procedure for the Automatic Recognition of Dysfluencies in the Speech of Children Who Stutter

Abstract

This program of work is intended to develop automatic recognition procedures to locate and assess stuttered dysfluencies. This and the preceding article focus on developing and testing recognizers for repetitions and prolongations in stuttered speech. The automatic recognizers classify the speech in two stages: In the first the speech is segmented and in the second the segments are categorized. The units segmented are words. The current article describes results for an automatic recognizer intended to classify words as fluent or containing a repetition or prolongation in a text read by children who stutter that contained the three types of words alone. Word segmentations are supplied and the classifier is an artificial neural network (ANN). Classification performance was assessed on material that was not used for training. Correct performance occurred when the ANN placed a word into the same category as the human judge whose material was used to train the ANNs. The best ANN correctly classified 95% of fluent, and 78% of dysfluent words in the test material.