A fuzzy approach to statistical models in speech and speaker recognition

A unified fuzzy approach to statistical models for speech and speaker recognition is presented. Since the expectation-maximisation (EM) algorithm is a powerful learning method for maximising the likelihood of the observed data in the presence of hidden variables, the fuzzy EM algorithm based on the fuzzy c-means algorithm is thereby established. From this fuzzy EM algorithm, the fuzzy algorithms for hidden Markov models, Gaussian mixture models, and vector quantisation are developed. The experimental results on T146 and ANDOSL speech data corpora for speech and speaker recognition show that the fuzzy approach is capable of achieving higher recognition accuracy.