英语口语机考评分系统中语种识别方法

Solution toLanguage Identification in CALL English Oral Test

文章阐述了基于计算机的英语口语教学与测试的现况和存在问题,试图用基于混合高斯模型(GMM)的语种识别系统来进行语种"拒识别",从而改变实际环境和个人因素这两个一直影响识别率提高的主要因素。根据英语口语教学与测试的实际运行情况,从混合高斯模型帧似然概率的统计特性出发,提出一种新的帧似然概率变换方法,即把帧概率按照一定规则变换成权值。在英语口语教学与测试实施过程中,语种识别时不依靠概率的绝对值,而是依靠帧概率在所有混合高斯模型中的相对位置来决定模型总得分。理论分析和实验结果表明,同传统方法相比,该方法能较好地提高语种识别率,从而起到了对其他语种的"拒识别"。

The paper firstly addresses the situation and problems in CALL english oral teaching and test,trying to adopt GMM-based language identification system in CALL english oral teaching and test,hence the refusal of identification of other languages.In the language identification system,the environment and individual characteristics are always the factors that influence the identification accuracy.A new conversion method based on likelihood probability of frames of the system,is presented in the paper,which analyzes the statistical specialties of likelihood probabilities of frames.The weights are got from the probabilities of frames according some rules so that the recognition result doesn‘t depend on the absolute value of the probability,but on relative positions of the frame probability in all other language models.Compared to traditional methods,the theoretical analysis and experimental results show that the method presented in the paper can better improve the identification accuracy as to refuse other languages identification.