GFCC-Based Robust Gender Detection

Abstract

Gender classification technique is a part of the signal processing comprises with feature extraction and behavioural gender modelling. Fundamental frequency and pitch are mostly used as feature for gender detection due to their unique characteristics in voice source. In this study, Gammatone Frequency Cepstral Coefficient (GFCC)-based robust gender classification method has been presented. This study was accomplished using speech samples from a text-dependent data set. The prototype gender behavioural modelling was done using Gaussian mixture model (GMM) to obtain better performance and only clean signal was used to train the model. The performance of the proposed method was tested under both clean and contaminated conditions. The clean signal was contaminated using nine different noises at a range of signal-to-noise ratios (SNRs) from 0 dB to 10 dB. The obtained performance showed the proposed method was very robust against noise and the average performance at 0 dB SNR was almost 100% for female and 92% for male irrespective to noises. So, it could be said the proposed method performance was almost noise invariant.