基于语音识别置信度的英语语言学习算法研究  

English Language Learning Based on Confidence Measures

作　　者：贲俊[1] 余小清[1] 万旺根[1] DanielP.K.Lun 

机构地区：[1]上海大学通信与信息工程学院,上海200072 [2]香港理工大学电子信息工程系多媒体信号处理中心

出　　处：《上海大学学报（自然科学版）》2002年第5期391-395,399,共6页Journal of Shanghai University:Natural Science Edition

基　　金：国家自然科学基金 (60 0 72 0 3 1);香港理工大学研究基金 (G-T499)资助项目

摘　　要：语音识别技术是实现一个语言学习系统的关键 .由于目前的英语学习软件并不能完全满足用户的要求 ,因此该文提出了将语音识别置信度引入语言学习的方法 .在讨论了置信度的基本原理和在语言学习中的作用之后 ,提出了一种新的算法并在此算法基础上最终建立了一个基于置信度技术的英语语言学习系统 .在 TIMIT语音库中的实验结果表明 ,语音识别置信度可以有效地提高语言学习系统的性能 ;当门限值设置于 - 0 .7和 - 1.0之间 ,虚警率 (false alarm)在 1.1%左右 ,漏报率 (false rejection)在 0 .6 5 %左右 ,并且在一定条件下可以有效地反映学习者发音的质量 ,基本上可以达到实用的要求 .Speech recognition is the key technology in language learning system. However, the present English learning programs can not satisfy user's expectation. In this paper, we introduce confidence measures into English language learning. The basic theory and the role of confidence measures in English language learning are discussed. We then propose a new algorithm, and develop an English language learning system based on confidence measures. Experiments are conducted based on TIMIT database and the results show that confidence measures can be utilized to effectively improve the performance of language learning system; FA (false accepted out of vocabulary words) is about 1.1%, and FR (false rejected for in vocabulary words) is about 0.65% when the threshold is set between -0.7 and -1. Moreover, in a certain condition the system can reflect the quality of learners' pronunciations, which basically satisfies the practical demand.

关 键 词：语音识别 置信度 英语语言学习算法 语言学习系统 计算机辅助发音训练 门限值 

分 类 号：H319[语言文字—英语] TN912.34[电子电信—通信与信息系统]