Neural-Response-Based Text-Dependent Speaker Identification Under Noisy Conditions

Abstract

Speaker identification is a technique of determining an unknown speaker’s identity and is very essential for security, crime investigation, and telephoning. In this study, a text-dependent speaker identification technique using the neural responses of a physiologically-based computational model of the auditory periphery is proposed. Neurograms were constructed from the responses of the auditory-nerve model to sentences of different speakers. The proposed features were then used to train and test the recognition system using the support vector machine and Gaussian mixture model classification techniques. The proposed method was tested on a text-dependent database in quiet and under noisy conditions for a range of signal-to-noise ratios. Although the performance of the proposed method in quiet was comparable to the performance of a traditional Mel frequency cepstral coefficients-based method and also to the result of a very recent Gammatone frequency cepstral coefficient-based system, the neural-response-based method showed a substantially better classification accuracy under noisy conditions. The proposed method could be extended to design a text-independent speaker identification and verification system.