Effect of performance parameters of SVM and k-NN on speech recognition for articulatory Handicapped people

Abstract

Speech Recognition is the biggest challenge in case of disordered speech, because of unavailability and diversity of database. In this paper, we use MFCC as feature extraction method as they provides speech features similar to the way how human hears and perceives sounds. For decision of predicted word k-NN and SVM classifiers are used. The fundamental target of this paper is to discover the execution of k-NN and SVM classifier. Classifier performance evaluated on various parameters. The database consists of different samples includes zero to ten digits spoken by different speakers who suffer from different types of speech disorders. Experimental results show that k-NN gives highest prediction accuracy than SVM.