文献类型：会议论文

英文题名：A speech recognition system based on multiple neural networks

作者：Lu, Bo[1]; Wu, Jing-Jing[1]; Wang, Yu[1]; Li, Jin-Ping[1]

第一作者：Lu, Bo

通讯作者：Li, J.-P.

机构：[1] College of Information, Beijing Union University, Beijing, China

第一机构：北京联合大学智慧城市学院

会议论文集：Proceedings - 2010 6th International Conference on Natural Computation, ICNC 2010

会议日期：10 August 2010 through 12 August 2010

会议地点：Yantai, Shandong

语种：英文

外文关键词：Neural networks - Spectrographs - Speech - Template matching

摘要：According to the visible characteristics of speech, a new speech recognition system design was proposed. It based on multiple neural networks. Firstly, Pulse Coupled Neural Network(PCNN) was input into the spectrogram for producing the corresponding time series icon as the feature parameters of speech. And then, the feature parameters was input into the Probabilistic Neural Networks(PNN) for training PNN to realizing speech recognition. Finally, the Probabilistic Neural Networks(PNN) was used of processing these feature parameters for realizing speech recognition. This way could solve the problem of template matching weak generalization of PCNN. As the simulation results shown, the higher speech recognition rate could be gotten by this way. ? 2010 IEEE.