Abstract: The paper proposes a novel feature based on an auditory periphery model for robust speaker recognition.The sub-band energies of theextracted auditory features are calculated using a Gammatone filterband insteadof commonly used triangle filter band.The center frequencies and bandwidthsare then determined according to the equivalent rectangular bandwidth (ERB) model.Moreover,weighting the Gammatone filter bank by analyzing contribution of short-time spectrum in different frequency sub-bands,and using the CMS method toremove the varibility of channels are also investigated.Simulation results withGaussian Mixture model indicate that the recognition accuracy is significantly improved by this auditoryfeature in the noisy environments for the text-indepentent speaker recognition,especially in low SNR environments.

Key words: auditory feature, Gammatone filter bank, sub-band weighting, speaker recognition