A Particle Swarm Optimization for Hidden Markov Model Training

A particle swarm optimization (PSO) is presented for training Hidden Markov Model (HMM) used in speech recognition. The PSO is designed to estimate optimal parameters of HMM. Some heuristic algorithms such as Baum-Welch algorithm are developed to optimize the model parameters to describe the training observation sequences. However, these methods are hill-climbing algorithms and easy to converge to local optimal solutions, which might deteriorate the speech recognition rate. A PSO-HMM training approach aimed at finding the global solution or better optimal solutions is proposed in this paper. Comparing the proposed approach with the Baum-Welch algorithm and genetic HMM training method, the experimental results show that it is superior to both the Baum-Welch and GA-HMM training methods