Robust gender classification using neural responses from the model of the auditory system

Human listeners are capable of extracting several information of the speaker such as personality, emotional state, gender, and age using features present in speech signal. The gender classification of a speaker based on his or her speech signal is crucial in telecommunication. This study proposes a gender classification technique using the neural responses of a physiologically-based computational model of the auditory periphery. Neurograms were created from the responses of the model auditory nerve to speech signals. Orthogonal moments were applied on the neurogram to extract features for classification using Gaussian mixture model. The performance of the proposed method was evaluated for eight different types of noise. The result showed a high accuracy for gender classification for both under quiet and noisy conditions. The proposed method could be used as a pre-processor in speaker verification system.