Conclusion

A fuzzy approach to statistical models in speech and speaker recognition has been presented in this paper.A family of fuzzy EM algorithms based on the FCM has been proposed and inciuding the EM algorithm itself as one of the fuzzy EM algorithms with the degree of fuzziness of one.We have also developed the fuzzy EM algorithm into the fuzzy HMM，the fuzzy GMM and shown that the fuzzy VQ，can be derived from the fuzzy EM.These new algorithms can be regarded as effectively fuzzy modifications for statistical models.Since this fuzzy approach is not complicated，it can be hopefully applied to other EM-type algorithms to develop into the corresponding fuzzy EM-type algorithms.