Healthy and Parkinson voices discrimination based on compensation/normalization cepstral features

Human speech is one of the most affected by a neurodegenerative disease. In this paper, a study case of acoustic features for persons affected by Parkinson disease is done in order to discriminate voices of respectively healthy individuals (HI) and persons with Parkinson (PWP). For this purpose, a set of Mel frequency cepstral coefficients is extracted from speech of healthy and Parkinson disease individuals. Promising performance of discrimination is reached (97 %) thanks to the use of compensation/normalization of features along with K-nearest neighbor classification.