Identification of Voice Utterance with Aging Factor Using the Method of MFCC Multichannel

TL;DR

This paper proposes a multichannel MFCC-based method to improve voice identification accuracy over long periods affected by aging, achieving up to 86% compatibility over 10 years.

Contribution

It introduces a multichannel MFCC feature extraction approach to better handle voice changes due to aging in speaker identification.

Findings

M5FB and M2FB models achieved 85% and 82% compatibility over 25 years.

M5FB model achieved 86% compatibility over 10 years.

The multichannel approach improves long-term voice identification accuracy.

Abstract

This research was conducted to develop a method to identify voice utterance. For voice utterance that encounters change caused by aging factor, with the interval of 10 to 25 years. The change of voice utterance influenced by aging factor might be extracted by MFCC (Mel Frequency Cepstrum Coefficient). However, the level of the compatibility of the feature may be dropped down to 55%. While the ones which do not encounter it may reach 95%. To improve the compatibility of the changing voice feature influenced by aging factor, then the method of the more specific feature extraction is developed: which is by separating the voice into several channels, suggested as MFCC multichannel, consisting of multichannel 5 filterbank (M5FB), multichannel 2 filterbank (M2FB) and multichannel 1 filterbank (M1FB). The result of the test shows that for model M5FB and M2FB have the highest score in the level of compatibility with 85% and 82% with 25 years interval. While model M5FB gets the highest score of 86% for 10 years time interval.