基于反转HMM的双层搜索策略及其应用  

Two-level Search Strategy Based on Reverse HMM and Its Applications

作　　者：王晓兰[1] 郭玲[1] 周献中[1] 黄志同[1] 

机构地区：[1]南京理工大学自动控制系,南京210094

出　　处：《计算机工程》2004年第3期61-62,121,共3页Computer Engineering

基　　金：国防跨行业基金资助项目

摘　　要：通过对连续发音的数字串识别实验的搜索过程进行分析,发现许多误识是由于声母段及声韵段叠信息丢失而造成的。因此结合汉语音节结构特点,通过引入了反转HMM,提出了双层搜索策略,较好地解决了单层搜索时音节分界点错位引起的声母段及声韵段叠信息丢失而造成的误识现象。In this paper the recognition process of continuous mandarin digit string speech is analyzed. The analysis results show that many recognition errors are caused by the loss of initial parts of syllable. Considering the special structure of mandarin syllable, RHMM is introduced. In the search process two-level search strategy based on RHMM is used. The improved algorithm works well in overcoming the recognition errors caused by the loss of initial parts of syllable. Experimental results confirm the effectiveness of the proposed method.

关 键 词：语音识别 反转隐马尔可夫模型 双层搜索策略 

分 类 号：TN912.34[电子电信—通信与信息系统]