A Robust Text Dependent Speaker Identification Using Neural Responses from the Model of the Auditory System

Abstract

Speaker recognition is considered as a behavioral biometric to identify speaker’s identity based on their voice features. In this study, a new speaker identification system is proposed using the neural responses at the level of the auditory nerve (AN). For this, a very well-developed physiological based computational model of auditory periphery is used to simulate the neural responses for a given speech. The output, in the form of synapse responses, is then analyzed for the feature extraction. Neurograms are constructed for a range of characteristic frequencies from the output responses. Features are then calculated from the neurogram to train the system. The same extracted features for a given speaker are then used to identify the speaker in the testing phase. To test the reliability of the proposed system, the model has been tested both in quiet and noisy conditions. The results show that, neural response-based speaker identification system can substitute the existing technology and thus improve the performance for application of remote authentication and security system.