A Speech Recognition System Based on Fuzzy Neural Network Optimized by Time Variant PSO

摘要

In order to overcome shortages of fuzzy neural network (FNN) and basic Particle Swarm Optimization (PSO) algorithm, the article proposes a novel method that the parameters of structure equivalent FNN (SEFNN) trained by Time Variant Particle Swarm Optimization (TVPSO) algorithm. TVPSO is made adaptive in nature by adaptively and dynamically changing its acceleration coefficients and its inertia weight with iterations and fitness value, which helps the algorithm to explore the search space more efficiently. The Parameters of SEFNN trained by TVPSO algorithm was used in speech recognition system which improve the ability of generalization and self-learning of FNN and is able to determine the fuzzy rule numbers according to the vocabulary to be recognized. The experimental results show that the SEFNN optimized by TVPSO for speech recognition system have faster convergence, higher recognition ratio and better robustness than SEFNN trained by PSO algorithm, FNN trained by BP algorithm.