中文摘要：

提出一种使用韵律模型的方法来改进计算机辅助语言学习中的发音质量评价性能．该方法在原有的单音子和三音子模型的基础上，添加了韵律因素．一方面，这使得对影响发音质量最大的元音音素的描述更为细致；另一方面，包含韵律模型的方法从一定程度上解决了使用母语训练库和非母语测试库之间的模型不匹配问题，为语言学习机在母语和非母语之间的交叉应用提供了可能．同时，这种改进的方法还为计算机辅助语言学习系统中的错误检测和反馈提供了很好的参考，更进一步增加了发音质量评价模块在整个学习机系统中的指导作用．试验中，使用WSJ（母语库）作为训练库，ESC（非母语库）作为测试库，基于带韵律的Mono—Phone模型得到的匹配分数，段长分数，感知分数融合结果与主观评价之间的最终相关性为0．839，比原有基本英语音素的方法的融合结果提高了0．08（0．753）．

英文摘要：

In this paper, we proposed an improved prosodic method to realize the pronunciation evaluation module for computer aided language learning（CALL）. It was realized on the base of perceptual and acoustic model,in which the prosodic characters were tied to the phone model. With this model,the phone models are more accurate to represent the speech character,also it weakens the mismatch between native and non-native articulation and provides profound chances for later application. At the same time, it can provide useful reference for error detection and feedback for later modules of the whole CALL system. So the instructional function of the pronunciation evaluation module to the language learning system is strengthened a lot. In our experiment, the correlation between fusion score and expert score was improved from 0. 72 to 0. 834（vowel score,consonant score,perceptual score and duration score） ,which corresponding to basic phone models and prosodic phone models separately.