Speaker recognition using neural responses from the model of the auditory system

Speaker recognition is a process of determining a person´s identity using features in speech signals. In this study, a new speaker recognition (identification and verifica-tion) system is proposed using the responses from a computational model of the auditory system. A neurogram (2D) was constructed from the responses of the model of auditory nerve fibers for a range of characteristic frequencies. The proposed neurogram based speaker recognition system was trained and tested using a Gaussian mixture model classification technique. The performance of the proposed method was evaluated for both clean speech and speech under noisy environment. The result of the proposed method was compared to a traditional speaker recognition technique, referred to as the mel-frequency cepstral coefficient method. The proposed method showed better performance than the traditional approach, especially under noisy conditions. The proposed method could be applied in security and voice recognition systems.