Training algorithm of hidden Markov model based on mixture of factor analysis

Combining mixture of factor analysis method with hidden Markov modeling techniques, a new statistical acoustic model was constructed: hidden Markov model based on mixture of factor analysis (HMM-MFA). The HMM-MFA models the correlation between the feature vector elements in speech signals. The training algorithm for HMM-MFA was studied, by generalizing Baum's auxiliary function into this framework and an associated objective function was built up. The training equations for estimating parameters of HMM-MFA was derived by Lagrange multiplier method. Simulation shows that the proposed algorithm is better than the traditional EM algorithm in speech recognition accuracy.