MFCC based recognition of repetitions and prolongations in stuttered speech using k-NN and LDA

Abstract

Stuttering is a speech disorder in which the normal flow of speech is disrupted by occurrences of dysfluencies, such as repetitions, interjection and so on. There are a high proportion of repetitions and prolongations in stuttered speech, usually at the beginning of sentences. Consequently, acoustic analysis can be used to classify the stuttered events. This paper describes particular stuttering events to be located as repetitions and prolongations in stuttered speech with feature extraction algorithm. The well known Mel Frequency Cepstral Coefficient (MFCC) feature extraction is implemented to test its effectiveness in recognizing prolongations and repetitions in a stuttered speech. In this work, two classifiers such as Linear Discriminant Analysis based classifier (LDA) and k-nearest neighbors (k-NN) are employed and k-fold cross-validation was applied to measure classifiers performances. The result of this work shows that the MFCC and classifiers (LDA and k-NN) can be used for recognition of repetitions and prolongations in stuttered speech with the average accuracy of 90%.