Speaker Identification Based On MFCC and IMFCC

There are the features of high recognition rate and strong power against noise for Mel-Frequency Cepstral Coefficients (MFCC) that modeled on the human auditory system compared with other features. However, due to the structure of its filter bank, it captures characteristics information more effectively in the lower frequency regions than in the higher regions. Thus there must be some information contained in the high frequency is lost. This work uses a new set of features by inverting the filter bank structure which can make up the drawback of MFCC. Considering the complementary relationship of the two features MFCC and IMFCC, an identification method which combines the decision result of two classifiers is presented. The experimental results show that IMFCC is feasible as the features of speaker identification. The performance of decision system has been improved by the method of combining multi-classifiers.