关键词: 模糊, 神经网络, 语音识别

Abstract: An improved T-S fuzzy neural network with four layers network structure is introduced.This algorithm adds a compensated factor interrelated with the input dimensions on the membership degree，which can be applied into the speech identification system.In the meanwhile it also solves the “regulation disaster” problems caused by the intensive input dimensions.The experiment results show the improved T-S fuzzy neural network is applied to speech recognition system，and the recognition ratio of this network is higher than RBF network，meanwhile obtains much better robustness.

Key words: fuzzy, neural network, speech recognition