Comparison of Feature Extraction for Speaker Identification System

Abstract

This paper compares the performance of speaker identification systems based on feature extraction methods. Fast Fourier Transform (FFT), Mel-Frequency Cepstral Coefficient (MFCC) and Discrete Wavelet Transform (DWT) are three of chosen feature extraction techniques used to test. These methods are applied to identify speakers by a word spoken. The system used Dynamic Time Warping (DTW) as classifier. Programming is done on MATLAB for training and testing. In this experiment, the combination of DWT and DTW gives better accuracy result than the other methods.