Speech stuttering assessment using sample entropy and Least Square Support Vector Machine

This work is intended to discuss the performance of sample entropy feature for the recognition of stuttered events. The data for the analysis is taken from the UCLASS database. Manual segmentation is performed to identify the stuttered events prior to the feature extraction process. Wavelet packet decomposition is performed, and the sample entropy features are extracted using three different filter banks, Bark scale, Mel scale and Erb scale. The extracted features are tested using Least Square Support Vector Machine (LS SVM) for the identification of repetition and prolongation. Ten fold cross validation method is used to ensure the reliability of the results. The experimental investigations reveal that the proposed method shows promising results in distinguishing between the two stuttering events.