Text-dependent speaker identification based on mutual information matching model

The Mutual Information Matching model (MIM) was proposed for speaker recognition based on the mutual information theory. Both of statistical and time-variant features of speech signal can be processed effectively, robustly and synchronously in MIM. It is presented a description of MIM principle and then evaluated its application to text-dependent speaker identification with comparison to other two typical models, DTW and GMM. The identification experiments on 30 speakers including 18 males and 12 females show that MIM model has better performance with the identification error rate of 1.33% if LPCC was used as feature parameters.