A Hybrid Speech Recognition Training Method for HMM Based on Genetic Algorithm and Baum Welch Algorithm

HMM describes the time-domain feature of speech signal by statistical modeling method. Classical training method Baum Welch algorithm could only obtains locally optimal solution, which might decrease the recognition rate, while an important character of genetic algorithm is global search, so we can get a globally optimal solution or at least sub-optimal solution. In this paper genetic algorithm was applied to the optimization of the initial value of B in Baum Welch algorithm. A hybrid training method that combined the traditional method with genetic algorithm was proposed. Experimental results showed that the method had both qualities of global search and rapid convergence and the resulting models were superior to those obtained with traditional methods.