基于HTK的语音识别语言模型设计及性能分析  

Design of language model of speech recognition based on HTK and its performance analysis

作　　者：张强[1] 陶宏才[1] 

机构地区：[1]西南交通大学信息科学与技术学院,四川成都610031

出　　处：《成都信息工程学院学报》2009年第2期142-146,共5页Journal of Chengdu University of Information Technology

摘　　要：在大词汇连续语音识别系统中,统计语言模型是该系统非常重要的一部分,对整个识别系统的识别效果产生重要的影响。分析了统计语言模型原理和相关技术,基于HTK设计了适合大词汇连续语音识别系统的统计语言模型,同时对该语言模型的复杂度、精确识别率进行分析,指出了影响统计语言模型性能的相关因素。In the Large Vocabulary Continuous Speech Recognition （LVCSR） the statistical language model plays an important role in the entire speech recognition system. The principle and rdated techniques of the statistical language model are analyzed and the model suitable for the large and continuous speech recognition system is designed by using the HTK. The complexity and accurate recognition rate of the model are analyzed. The related factors affecting the performance of the model are pointed out.

关 键 词：语音识别 统计语言模型 HTK N-GRAM模型 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]