Diagnosis of individuals who stutter is carried out by Speech-Language Pathologist (SLP). In order to calculate the rate of speech stuttering severity, SLPs traditionally count the number of speech dysfluencies and divide it by total number of spoken words. As can be inferred, the procedure in time consuming, …show more content…
In fact, there is low agreement on the true rate of stuttering severity of a client among different SLPs due to slightly different definitions of stuttering proposed by clinicians and their mistake in counting the dysfluencies. The mentioned problems can be alleviated by an accurate system to detect types, durations and numbers of dysfluent moments in continuous speech automatically.
The task of automatic dysfluency classification was started with a simplified framework in which the dysfluent moments were segmented manually. Authors used speech envelope parameters as feature set and artificial neural network (ANN) as classifier to recognize dysfluency classes. This framework has been adopted by many other researchers to test the other features such as MFCC, LPCC and PLP and other classifiers such as HMM, KNN, SVM and