利用HMM嵌入训练方法建立汉语电话连续语音声学模型  

Modeling For Chinese Continuous Telephone Speech Using HMM's Embedded-Training Method

作　　者：张宗红[1] 陈愉[1] 冯哲[1] 邵央[1] 李宗葛[1] 

机构地区：[1]复旦大学计算机科学系,上海200433

出　　处：《计算机工程与应用》2000年第6期36-38,共3页Computer Engineering and Applications

摘　　要：文章介绍了用HMM嵌入训练方法来建立连续语音的声学模型，并对基于音素的HMM和基于音节的HMM进行了比较，并以此为基础建立了一个实用的银行电话语音服务系统。This paper describes the building of a set of Chinese acoustic models which used to recognize telephone utterances by embedded-training method, and compares phoneme-based HMMs with syllable-based HMMs. Based on these models, a banking service system is constructed.

关 键 词：嵌入训练 电话语音识别 连续语音 声学模型 HMM 

分 类 号：TN912.34[电子电信—通信与信息系统]