Pitch-based cepstral features for gender classification in noisy environments

Abstract

Pitch based cepstral features are proposed for a novel gender classification system. A clean and noisy database of 100 speakers, 59 males and 41 females has been used for performing the experiments. Training and testing was performed using Gaussian Mixture Models (GMM) in clean and noisy environments. The proposed, Pitch Based Linear Prediction Cepstral Coefficients (PLPCC) and Pitch Based Mel Frequency Cepstral Coefficients (PMFCC) has shown a maximum of 12.12% and 15.62% increment in performance over Linear Prediction Cepstral Coefficients (LPCC) and Mel Frequency Cepstral Coefficients (MFCC) respectively.