Abstract: HMM is good at dealing with sequential inputs, while SVM shows superior performance in classification. Furthermore, the former approach usually provides an intra-class measure while the latter proposes inter-class difference. Since these two classifiers use different criteria, they can be combined to yield an ideal one. The output of support vector machines is converted into the form of posterior probability which is computed by the combined use of sigmoid function and Gaussian model, it acts as a probability evaluator in the hidden states of hidden Markov models. Experiments on speaker verification show that this hybrid model achieves better performance than continuous density hidden Markov models.