Voice as a Robust Biometrics

Robust voice based features for biometric authentication in noisy environments are proposed. The proposed processing includes gamma tone auditory bandpass filtering of speech signal, rectification, and compression to model the effects of the auditory system periphery. Three features are extracted by applying independent component analysis to the frequency, cepstral and auto-correlogram domains of the compressed output signals respectively. A specially prepared noisy speech corpus was used to gauge the performance of the proposed features on a speaker identification system. Experimental results show that these features can well denote the distribution of speakers and are robust to background noises compared with the traditional features, such as LPCC, MFCC and PLP. Among the proposed features, the feature extracted in auto-correlogram domain achieves the best identification performance in noisy-mismatched environments.