Processing and Analysis of Human Voice for Assessment of Parkinson Disease

Acoustic analysis of voice is a non invasive, reliable, easy to use and cost effective method in detecting parkinson disease. Voice deviation from normal one is the earliest indicator of parkinson disease. Voice data of sustained phonation is collected from 25 healthy and 22 parkinson subjects. The voice database is analyzed and acoustic features are extracted. Two new parameters ECP (energy between consecutive peaks) and ASR (average slew rate) are defined. The values of these parameters show variation among two groups. A row vector is prepared using these parameters and fed to the classifiers. ANN (artificial neural network) with Cascade, Feedforwad and Elman back prop functions, SVM (support vector machine), Discriminant analysis with linear and diaglinear functions are used as classifiers and their performances are compared. SVM has been tested to be the best one and gives the accuracy of 86%. Performances of classifiers are evaluated in terms of sensitivity, specificity and accuracy.