A method based on septrogram and pitch for biometric authentication

The performance of speaker recognition (SR) system may deteriorate seriously due to the different speaking conditions, especially mismatch recording channels. In this paper we propose an effective approach to conduct the SR task more reliable. This method is based on spectrogram and pitch, in which spectrogram (visual displays) of speech signal uttered from different persons is used as the identified feature. We illustrate the pith (fundamental frequency) is also important for speaker recognition. Furthermore, we address some fundamental pitch tracking techniques, based on time autocorrelation and average magnitude difference function.