Voice analysis for detecting patients with Parkinson's disease using the hybridization of the best acoustic features

Parkinson's disease (PD) is a neurodegenerative disorder of unknown etiology. It causes, during its course, vocal impairment in approximately 90% of patients. PD patients suffer from hypokinetic dysarthria, which manifests in all aspects of speech production: respiration, phonation, articulation, nasality and prosody. To evaluate these, clinicians have adopted perceptual methods, based on acoustic cues, to distinguish different disease states. In order to improve these evaluations, we used a variety of voice samples comprising the numbers from 1 to 10, four rhymed sentences, nine Turkish words plus the sustained vowels "a", "o", and "u". Samples were collected from 40 people, 20 with PD. We used the method of Leave- One-Subject-Out (LOSO) validation with a K Nearest Neighbor (k-NN) classifier with its different types of kernels, (i.e.; RBF, Linear, polynomial and MLP). The best result obtained was 82.5% using two diffirent voice samples; 1- the 4 th acoustic features along with the 17 th voice samples; 2- The 3 th and the 5 th acoustic features along with 20 th voice sample.