摘要
连续语音识别中，由于各声学单元之间没有明显的间隔或声学标志，匹配和搜索算法成为影响系统实时性能的核心问题．基于汉语语音的特点，提出了一种改进的Ｎ－Ｂｅｓｔ搜索算法．该算法大大降低了传统Ｎ－Ｂｅｓｔ程序搜索空间而不至于失去最优解，有利于算法的实时实现．
In continuous speech recognition, search strategy is of great importance in determining the practicability of a recognizer because there is no explicit interval or delimiter between acoustic units. This paper proposes a fast search matching algorithm based on the features of Chinese syllables for finding N Best hypotheses, which can greatly reduce the search space without loosing the optimal path, and remarkably improve the N Best search algorithm in the performance of implementation in real time.
出处
《西安电子科技大学学报》
EI
CAS
CSCD
北大核心
1999年第1期84-88,共5页
Journal of Xidian University
关键词
汉族语音识别
N-Best算法
连续语音识别
Chinese speech recognition N Best algorithm continuous speech recognition search algorithm
