Abstract: A complete feature corpus as speaker model and a evaluation algorithm of mutual information for text-independent speaker identification are proposed. The speaker model is trained by a clustering algorithm in feature vector space using speech samples with various representative pronunciation characteristics of the speaker. The evaluation algorithm is used to calculate the likelihood between input speech and the models in distance and information space, maximum mutual information decision rule is used to decide the identity of speaker. Experiments on performance analysis with comparison to GMM (Gaussian Mixture Model) method according to linear predictive cepstrum and Mel-fequency cepstrum parameters show the proposed model and evaluation algorithm is quite effective.