Abstract

Voice disorder classifications have developed more and more momentum now days because of complication in conventional methods. speech disorder diseases creates voice problem hence speech signal can work as useful tool to diagnose voice disorders. In this research work, normal & abnormal speech signals are taken & a system is designed to classify patients of chordectomy, laryngitis, laryngeal paralysis, psychogenic dysphonia, vocal cord cancer from normal. Speech signals are first preprocessed. The preprocessed signal is used for spectral analysis with which normal & abnormal speech signals are differentiated.Various features are extracted and after selecting relevant and efficient feature, these features are given to various artificial neural networks and SVM. The neural network used are MLP,GFF,Modular network. The accuracy for these networks is 60.3,65.97,52.57 respectively .Support vector machine was found to be optimum classifier with classification of 92.26% and the percentage correct dtermination for various disease like chordectomy, laryngitis, laryngeal paralysis, psychogenic dysphonia, vocal cord cancer and normal using SVM is 100, 100 , 94.28, 95.34, 81.25 and 84.9 respectively

Paper Details