文献类型：期刊文献

中文题名：基于神经网络的汉语孤立词语音识别

英文题名：Neural Networks Based Speech Recognition of Isolated Chinese Word

作者：朱淑琴[1];魏威[1,2]

第一作者：朱淑琴

机构：[1]北京联合大学师范学院;[2]中国人民大学信息学院

第一机构：北京联合大学师范学院

年份：2012

期号：9

起止页码：447-448

中文期刊名：微计算机信息

外文期刊名：Microcomputer Information

基金：基金颁发部门:北京联合大学;项目名称:基于本体的概念相似度计算;编号:ZK201004X;基金申请人:魏威

语种：中文

中文关键词：规整网络;人工神经网络;语音识别

外文关键词：alignment network; artificial neural networks ; speech recognition

摘要：研究基于神经网络的汉语孤立词语音识别问题,神经网络通常是针对静态模式而设计,输入结构是固定的,语音信号是一个时变信号,发音时音节的长短不可能完全相同,将人工神经网络用于语音识别时需要对其做一些必要的修正。本文将语音特征参数序列通过规整网络转换为状态转移矩阵。状态转移矩阵维数固定,反映语音时变特性。从而很好的解决了神经网络动态模式识别问题,实现了基于神经网络的孤立词语音识别。实验结果表明该系统具有良好性能。
Neural Networks based speech recognition of isolated Chinese word is discussed in the paper.Neural Networks is usually designed for static pattern which input structure is fixed.Speech signal is time-varying which lead to the length of the syllable pronunciation cannot be exactly the same.The application of Neural Networks in speech recognition needs some necessary correction.An alignment network used to extract fixed dimension feature vectors from input speech signal is proposed in this paper to address temporal varying problem in speech recognition.Consequently the dynamic pattern recognition using neural networks is solved.We also implement a high performance Neural-Networks-Based isolated words recognition system.Experimental results show the efficiency of the new algorithm in speech recognition.