基于DDBHMM的维吾尔语语音声学层识别系统研究  

Research on Uyghur Speech Acoustic Recognition System Based on DDBHMM

作　　者：吾守尔.斯拉木 刘俊[1] 王飞飞[1] 

机构地区：[1]新疆大学信息科学与工程学院,乌鲁木齐830046

出　　处：《新疆大学学报（自然科学版）》2010年第4期381-386,共6页Journal of Xinjiang University(Natural Science Edition)

基　　金：国家自然科学基金项目资助(批准号60762006)

摘　　要：采用基于段长分布的非齐次隐马尔可夫模型(DDBHMM)进行维吾尔语声学建模。在新语料下由于总词汇量的成倍增加导致识别时间倍增,为缩短识别时间将耗时最长的概率计算部分采用多线程机制优化了识别模块,同时加入了端点检测进行控制,并相继设计了录音模块、特征提取模块、波形显示及结果输出显示模块等,对这些模块进行集成界面化后产生了一个基于DDBHMM的维吾尔语连续语音声学层实时识别系统,并对系统进行了测试及验证.The acoustic modeling is established on the Duration Distribution Based Markov Model （DDBHMM）. But the recognition rate slowed down obviously for the vocabulary＇s doubly incensement, in view of this problem, we used multithreading in the longest time-consume compute part to optimize the recognition. Moreover, the vertex examination is under consideration, then the record module, feature- extraction module, wave-show module and output-show module were designed, finally, we integrated all the modules with interface, and developed an independent Uyghur continuous speech real-time recognition system based on DDBHMM, related tests have been done on the system.

关 键 词：语音识别 维吾尔语 DDBHMM 多线程 

分 类 号：TN912.34[电子电信—通信与信息系统]