摘要
提出了一种隐马尔可夫模型（ＨＭＭ）和径向基函数神经网络（ＲＢＦ）相结合的语音识别新方法。该方法首先利用ＨＭＭ生成最佳语音状态序列，然后用函数逼近技术产生对最佳状态序列进行时间规正，最后通过ＲＢＦ神经网络进行分类识别。理论和实验结果表明，该系统比ＨＭＭ具有更好的识别效果，特别对提高易混淆词的识别性能尤为显著。
Presents a new hybrid framework of hidden Markov models (HMM) and radial basis function (RBF) neural networks for speech recognition. Here, the HMM is employed to produce a best speech state sequence which is warped to a fixed dimension vector and the RBF neural network is used as classifier. The theoretical analysis and experimental results show that the new hybrid system works better than HMM especially in recognition of highly confusable words.
出处
《数据采集与处理》
CSCD
1999年第2期153-156,共4页
Journal of Data Acquisition and Processing
基金
国家攀登计划认知科学(神经网络)重大关键项目
关键词
神经网络
语音识别
隐马尔可夫模型
径向基函数
neural networks
speech recognition
hidden Markov models
radial basis function
