基于混合模型HMM/RBF的数字语音识别  

MDSR Based on Hybrid HMM/ RBF

作　　者：王朋[1] 陈树中[1] 

机构地区：[1]华东师范大学计算机系,上海200062

出　　处：《计算机工程》2002年第12期136-138,共3页Computer Engineering

摘　　要：提出一种离散隐马尔科夫模型(hidden Markov model，HMM)和径向基函数(radial basis function,RBF)神经网络相结合应用于汉语数码语音识别(Mmandarin Ddigit Speech Recognition,MDSR)的方法。同时采用了一系列改进方法，使汉语数码语音的识别率达到了99.7%。This paper presents a new hybrid framework of hidden Markov model (HMM )and radial basis function (RBF) neural network for Chinese Mandarin digit speech recognition.Here, the HMM is employed to produce a best speech state sequence, which is warped to a fixed dimension vector and neural network is used as classifier.Results show that the new hybrid system works better than others.

关 键 词：混合模型 HMM RBF 数字语音识别 数码语音识别 隐马尔科夫模型 径向基函数 人工神经网络 

分 类 号：TN912.34[电子电信—通信与信息系统]