Hybridization of best acoustic cues for detecting persons with Parkinson's disease

Parkinson's disease (PD) is a degenerative disorder of unknown etiology. It causes vocal impairment in approximately 90% of patients. In order to improve the assessment of speech disorders in patients with PD, and because objective acoustic analysis methods do not always yield correct diagnosis, we present in this paper a method of hybridization of acoustic parameters that gives improved diagnosis results. The features were selected according to the pathological thresholds defined by the Multi-Dimensional voice program (MDPV). Extracted acoustic features were fed into k-nearest neighbor (k-NN) and support vector machines (SVM), which were trained to classify the voice as pathological or normal. In this work, we collected a variety of voice samples from 14 patients with PD (7 female, 7 male) and 6 healthy subjects (2 female, 4 male). The best classification accuracy achieved was 95%.