Neural network based detection of voice disorders using energy spectrum and equal-loudness contours

Abstract

Impairment of vocal function can have a major impact on the quality of life, severely limiting communication at work and affecting all social aspect of daily life. In the recent years, voice disease are increasing dramatically due to unhealthy social habits and voice abuse. Acoustic analysis is a non-invasive technique to detect and diagnose the voice disorders. In this paper, a simple feature extraction method based on band energy spectrum and weighing factor of its center frequency derived from Equal-loudness contours is proposed. A simple Elman recurrent network models is developed for testing the proposed features. The simulation results indicate that the proposed algorithm can be distinguish the voice as pathological or non-pathological voice and provides the mean classification accuracy of above 90%. The proposed method has been potential for diagnosing the voice disorders.