Age Recognition Based on GMM and DBN with Using Wavelet Packet Mel-Frequency Cepstrum Coefficient

Abstract

A new multi-resolution feature extraction algorithm is proposed for the speaker and text independent age recognition. The new feature called Wavelet Packet Mel-Frequency Cepstrum Coefficient (WPMFC) takes more speakers personality characteristics like age compared with common feature such as MFCC. The speaker age is divided into four age groups such as children, youths, adult and older, and totally eight Gaussian Mixture Models (GMM) and Deep Belief Network (DBN) models are trained for each age group and gender. The performance of the age recognition systems with using WPMFC as the feature parameter are analyzed based on GMM and DBN. The experimental results show that the performance of the age recognition system based on GMM is more effective, the average recognition rate of outset speaker age reaches 65.17% which is superior to DBN approach for the same speech corpus.