DHMM speech recognition algorithm based on immune particle swarm vector quantization

Abstract

This article presents a novel Immune Particle Swarm Optimization (IPSO), which combines the artificial immune system methods like immunologic memory, immunologic selection and vaccination together, by making reference to the self adjusting mechanism derived from biological immune system. IPSO as a method of Vector Quantization applied to the Discrete Hidden Markov Model (DHMM) and proposes IPSO-DHMM speech recognition algorithm. Each particle represents a codebook in the algorithm. The experiments using IPSO vector quantization algorithm get optimal codebook. Finally it enters the DHMM speech recognition system to train and recognize. The experimental results show that the IPSO-DHMM speech recognition system has faster convergence, higher recognition ratio and better robustness than the PSO-DHMM algorithm. © 2011 Springer-Verlag.