The Speaker Recognition System Based on the Dynamic MFCC
Zhi Dong · 2005
In this paper a new speaker recognition algorithm based on the dynamic features of Mel Frequency Cepstral Coefficients(MFCC) is presented. The algorithm can dynamically give a set of Mel--filters which are designed according to the results of pitch detection. The Mel--filters are then used to extract a set of MFCC which is for identification of a speaker. This paper also addressed the problem of classifier design based on the Gaussian mixture model (GMM). A clustering based method for selection of optimal order number and initial estimations of the parameters of GMM is presented. The experimental results show that the method presented in this paper can obtain a good performance in a real environment.