Speaker Recognition System Using MFCC and HMM Methods

Abstract

Speaker recognition is a technology used to identify a person's identity based on their voice characteristics. This research aims to develop a speaker recognition system using the Mel Frequency Cepstral Coefficients (MFCC) method for voice feature extraction and the Hidden Markov Model (HMM) for speaker classification. Voice data was collected from 30 speakers in a total of 1500 voice samples. We test the data using the HMM model with five state configurations after preprocessing and feature extraction using MFCC. The test results showed that the accuracy of the training data ranged from 89.50% to 95.67%, while the accuracy of the test data was in the range of 83.63% to 89.46%. By conducting a rigorous evaluation, it can demonstrate superior performance in recognizing speakers with a high degree of accuracy with a combination of MFCC and HMM.