The Separation of Multi-Class Pathological Speech Signals Related to Vocal Cords Disorders Using Adaptation Wavelet Transform Based on Lifting Scheme

Abstract

Abstract. Achievement to a non-invasive method to properly diagnose the diseases is a significant subject in domain of speech processing. The aim of this paper is to apply non-invasive methods to do diagnosis, provide preventive strategies and plan for treatment aids and treatment. Regarding the speech disorders based on wavelet features of common wavelet although have relatively proper performance, it is expected that design optimizations based on the features of speech signal and classifier performance lead to improvement of results. To design the adaptive wavelet transform, the parameters of lifting scheme generating bi-orthogonal wavelet are initially applied and then they are optimized through genetic algorithm and classification performance of Support Vector Machine. The result separation of normal and pathological signals provides an accuracy of 100 percent. Also, the result of two-class and three-class separation of six disorders using adaptation wavelet based on lifting scheme which indicative the advantage of suggested method with other mother wavelet.