Speaker recognition using mel-frequency cepstrum coefficients and sum square error

Abstract

The Method of Mel-Frequency Cepstral Coefficients Vector Quantization (MFCC-VQ) can be used in the speaker verification system. The process of feature extraction of speech signal using Mel Frequency Cepstral Coefficients (MFCC) vectors will produce acoustic speech signal. Vector quantization (VQ) is used to form the specific acoustic vector for each speaker. The introduction or verification, Sum Square Error is used to match unidentified speakers with speakers in filebase by the smallest error. In this research, the system is used to verify the speaker, namely red, blue, and green in Indonesian. This system has been tested by comparing the success rates between sound source speaker verification are used as filebase and modeling to the sound source said that is not used as filebase. From 20 times the pronunciation of each test the percentage of success obtained a good speaker verification. On testing the speaker with the same as filebase, the average percentage of verification success was 70%, while testing the speaker that are not used as filebase obtain an average percentage of 83.3% successful verification.