Speech Fluency Measurement of Patients with Parkinson’s Disease by Forward-Backward Divergence Segmentation

Human speech is a complex process controlled by numerous factors. Diseases, such as Parkinson’s disease also affect the background neuronal activity of a patient, thus resulting audible and perceivable change in speech. In the present study speech fluency was measured by Forward-Backward Divergence Segmentation algorithm in the case of speakers with Parkinson’s disease and was compared to healthy speakers. FBDS algorithm yields speech fluency measures without the need of language dependent phoneme-level segmentation. Two types of continuous speech were examined separately: read text and short monologue by FBDS-derived features. The results indicate that the calculated features have significant differences between the two groups. Classification tests showed that the features can be used to distinguish the two groups to a certain degree. k-nearest neighbor method with Euclidean distance measure and support vector machines were used to obtain result with more than 85% accuracy, 92% sensitivity and 69% specificity. The features calculated in the present study can also be used to extend existing features sets by giving speech fluency measurements without the need of language dependent segmentation preprocessing.