Feature Vector Selection of Fusion of MFCC and SMRT Coefficients for SVM Classifier Based Speech Recognition System

Automatic speech recognition system involves two phases namely training and matching. Features extracted from known speech signals are used to train the system and is matched with the features generated from unknown speech. Selection of good technique for feature extraction and feature matching has a prominent effect in speech recognition. Feature vector selection, from Mel Frequency Cepstral Coefficients (MFCC) and Sequency based Mapped Real Transform (SMRT) coefficients, is presented in this work. Linear kernel based Support Vector Machine (SVM) is used as the classifier for recognizing the speech. The results show that fusion of transform coefficients increases the recognition rate and also the efficiency of the proposed system.