Pitch in Speaker Recognition

Abstract

In order to improve the speaker recognition accuracy, the pitch is applied to GMM-based speaker recognition (SR). The circular average magnitude difference function (CAMDF) method is used to extract the pitch. An endpoint detection method based on the pitch is proposed. The following four features are selected as the features of the SR: the mel-frequency cepstral coefficient (MFCC) based on the pitch, the pitch contour, the pitch first-order difference and the pitch changed rate. Experimental results show that the recognition rate using proposed endpoint detection method is improved 20% than that using the conventional method. The recognition rate of the proposed system using the selected four features is improved 5% than that of the speaker recognition system using the MFCC parameters only.