Audio-Visual Person Authentication Using Speech and Ear Images

This paper proposes a multimodal, biometric person authentication method using speech and ear images to attempt to improve the performance in mobile environments. It is well known that the performance of person authentication using only speech is deteriorated by acoustic noises and feature changes with time. Since the ear shape of each person does not change over time, integrating its image with speech information increases robustness of person authentication. Experiments are conducted using audio-visual database collected from 38 male speakers at five sessions over a half year period. Speech data are contaminated with white noise at various SNR conditions. Experimental results show that the authentication performance is improved by combining the ear image with speech in every SNR condition.