Robust Assessment of Dysarthrophonic Voice with RASTA-PLP Features: A Nonlinear Spectral Measures

Abstract

This paper presents an artificial intelligence based speech signal processing technique to identify dysarthrophonic voice with relative spectral-perceptual linear prediction (RASTA-PLP) features. Dysarthria is a neural motor speech disorder caused by muscular weakness. Voice analysis of dysarthrophonic patients is challenging as this disease has multidimensional effects on the human voice generation system. Conventional spectral analysis is unable to accurately characterize the pathology associated with nonlinear dynamicity of human voice. This work investigates the suitability of RASTA-PLP features excerpted from speech signals to identify dysarthrophonic patients. The speech samples of healthy and dysarthrophonic patients are collected from the Saarbrücken Voice Database (SVD). Several machine learning and Artificial neural network (ANN) based algorithms are developed to evaluate the classification performance of the proposed system. The designed system can achieve excellent performance in terms of accuracy (100%) considering female and male subjects separately.