The application of gene cloning in training of hidden Markov models

The classical Baum Welch algorithm for Hidden Markov Models (HMM) training is easily trapped in a local optimum. To solve this problem, we introduce the ideal of gene cloning and propose an algorithm which is based on gene cloning and the Baum Welch algorithm. When we obtaine the local optimum matrix B by iterative calculation of Baum Welch algorithm, the matrix B is deemed as an individual and operated on by gene cloning. At last, we select the best individual so as to further optimize the Hidden Markov Models(HMM). Experimental results showed that the speech recognition rate was improved under certain conditions.