基于音素的发音质量评价算法

面对广大的外语学习者,计算机辅助语言学习系统已经成为一种最佳的口语学习方式。该文提出了一种新的应用于计算机辅助语言学习系统的面向英语学习人群的发音质量评价算法,名为PASS(phone-basedautomaticscoreforl2speechquality)。PASS算法以基于隐含Markov模型的语音识别和口音自适应技术为基础,考察了音素发音的准确性和流利性信息,定义了音素级的发音质量分数,从而可以综合得到整句的评分结果。在实验室自行采集和精细标注的非母语语音库上与其他评分算法进行比较实验,PASS与专家评分的句子级相关性达到了0.66,优于其他算法。目前PASS算法已经被成功地应用于清华大学出版社的互动式语言学习系统中。

Phone-based pronunciation quality assessment algorithm

This paper presents an algorithm which can provide pronunciation quality assessments to foreign language learners in Computer-Assisted Language Learning (CALL) systems. The approach is based on non-native speech recognition and accent adaptation techniques using hidden Markov models. Confidence measures at the phoneme level are calculated to check the phone accuracy and speech fluency. The sentence pronunciation score is obtained by analyzing the phone quality results. The algorithm was evaluated using a corpus of non-native speech. Test results using this database show that the approach outperforms other assessment algorithms on correlations with expert scores at the sentence level. The system has been successfully adopted by the Tsinghua University Press interactive language learning system.