Enhancing Speaker Verification System using MFCC and HMM Method

Abstract

Speaker verification is a system to identify a person's identity using a person's characteristic data. In this study, we utilize the Mel Frequency Cepstral Coefficients (MFCC) to extract voice data characteristics with several stages including pre-emphasis, frame blocking, windowing, Fast Fourier Transform (FFT), Mel Frequency Wrapping, and discrete Cosine Transform (DCT) and employ the Hidden Markov Model (HMM) to classify voice data. This study uses 30 speaker voice data samples with 1500 speaker voice data samples and conducts preprocessing and feature extraction using several state configurations. The test results can harvest values from 40% to 100%, with the most accurate voice verification accuracy found in state 4 with 100% accuracy. The test results show that a combination of MFCC and HMM methods can be an accurate approach to the speaker verification process in real applications.