Classification of three pathological voices based on specific features groups using support vector machine

<span>Determining and classifying pathological human sounds are still an interesting area of research in the field speech processing. This paper explores different methods voice features extraction, namely: Mel frequency cepstral coefficients (MFCCs), zero-crossing rate (ZCR) discrete wavelet transform (DWT). A comparison is made between these order to identify their ability any input sound as a normal or voices using support vector machine (SVM). Firstly, signal processed filtered, then vocal extracted proposed finally six groups used classify data healthy, hyperkinetic dysphonia, hypokinetic reflux laryngitis separate classification processes. The results reach 100% accuracy MFCC kurtosis feature group. While other accuracies range between~60% to~97%. Wavelet provide very good with common like ZCR features. aims improve diagnosis disorders without need for surgical interventions endoscopic procedures which consumes time burden patients. Also, extraction offers reference further researches area.</span>