K-means nearest neighbor classifier for voice pathology

Abstract

The noninvasive acoustical analysis of normal and pathological voices help speech specialists to perform accurate diagnose of diseases. Pathological voices show higher vocal noise level due to malfunctioning of vocal cords. Addition of noise component in speech has found to change the spectral properties. In this study, we show the use of energy spectrum which is obtained from 21-channel filter-bank outputs, for the classification of pathological voices. A simple k-means nearest neighbor classifier based on spectral distance is designed and a good classification results have been reported.