The purpose of this revised proposal is to develop a reliable method for measuring chronic stuttering in the continuous speech of persons with this speech disorder. previous research has shown repeatedly that trained and untrained judges have exceedingly poor levels of agreement in identifying stuttering events. Indeed, there is evidence that trained judges in different treatment/research centers show dramatic differences even in the total number of stuttering events that they count on identical recordings of stutterers. The studies described for this proposal seek to solve this problem by investigating procedures that might lead via the use of artificial neural networks (ANN) to identify intervals of speech that reliable judges agree contain stuttering. These studies will have the following purposes: (1) to investigate judge characteristics and judgement methods that influence levels of interjudge agreement for intervals of stuttered and nonstuttered speech; (2) to evaluate the generality of these findings across different stuttering research centers; (3) to investigate procedures that might improve interjudge agreement for stuttering judgments above levels obtained within the two preceding studies; (4) to investigate the use of selected ANN's for identifying intervals of speech that judges reliably identify as stuttered and nonstuttered, plus compare ANN identified stutterings with those achieved by skilled judges.
