英语篇章朗读质量的自动评分  

Automatic scoring of English passage reading quality

作　　者：蒋同海[1] 张俊博[2] 潘复平[2] 颜永红[2] 

机构地区：[1]中国科学院新疆理化技术研究所,乌鲁木齐830011 [2]中国科学院声学研究所语言声学与内容理解重点实验室,北京100190

出　　处：《应用声学》2011年第6期418-426,共9页Journal of Applied Acoustics

基　　金：国家自然科学基金(No.10925419;90920302;10874203;60875014;61072124;11074275)经费资助项目

摘　　要：本文研究了英语篇章朗读的计算机自动评分。本文根据人工评分的角度和准则,用语音识别技术分析语音,提取一系列评价特征,包括朗读完整度特征、发音准确度特征、流利度特征,然后通过SVM回归把这些评价特征映射为质量分数。在对4000名中学生的英语水平自动测试中,用3200名学生的人工评分训练系统,对其余800名学生的机器自动测试取得分差为1.18的良好结果,而专家评分与参考评分的平均分差为1.31。实验表明该项技术已达到实用化水平。In this paper we studied the computer automatic scoring for English discourse oral reading. According to the view and guidelines of manual scoring, we analyzed the voices from oral reading with speech recognition technology, and extracted the series of features including reading completeness features, pronunciation accuracy features and fluency features for evaluation. We mapped these features to scores by SVM regression. In the testing of English discourse oral reading for 4000 middle school students, in which the materials of 3200 students were used to train and the rest of 800 students to test, we got a good result that the difference between average machine score and reference score is 1.18, while the difference between average human score and reference score is 1.31. The experience result shows this system can be used in practice.

关 键 词：自动评分 发音质量评估 语音识别 

分 类 号：TP311.1[自动化与计算机技术—计算机软件与理论]