Speech Recognition by Dynamic Time Warping Assisted SVM Classifier

Abstract

Speech recognition using sustenance vector machine assisted by Dynamic time warping (DTW) method is proposed. The input training datas are collected from 40 speakers for five unique words. Every one of the information was gathered in a profoundly acoustic and commotion confirmation condition. Mel recurrence cepstrum coefficients (MFCC's) are represented as constant property of the signal. First and second derivatives of MFCC are used for dynamic properties. Subsequent to deciding element vectors, an adjusted DTW technique is proposed for highlight coordinating. Support Vector Machine (SVM) as well as Radial basis function (RBF) are used to categorize. The model is tried for multiple speakers and a good detection rate is obtained.