论文部分内容阅读
从整体上介绍了汉语连续语音识别系统的基本原理 ,并重点对声学和语言两个层面的建模与搜索策略进行了分析 .在对传统帧同步搜索算法进行研究的基础上 ,提出了基于统计知识的帧同步搜索算法 SKB- FSS.它包含了三个主要的功能层次 :基于归并的音节切分自动机产生确定的搜索边界点 ,由统计得到的差分状态驻留信息控制搜索过程中的状态转移 ,利用词搜索树控制音节候选的扩展规模并根据动态前向预测的方法进行合理而及时的路径剪枝 .实验结果验证了该搜索策略的有效性 .
This paper introduces the basic principle of Chinese continuous speech recognition system as a whole, and focuses on the modeling and search strategies of both the acoustic and the language.On the basis of studying the traditional frame synchronization search algorithm, It consists of three main function levels: the merged syllable cut automaton generates a certain search boundary point, and the state information of the difference state resides in the control of the search process Transfer and use the word search tree to control the expansion scale of syllable candidates and make rational and timely path pruning according to the dynamic forward prediction method.The experimental results verify the effectiveness of the search strategy.