A evaluating model of english pronunciation for Chinese students
Guimin Huang, Jing Ye, Yan Guang Shen, Ya Zhou · 2017
This paper proposes an innovative model to evaluate English pronunciation for Chinese students. We use different acoustic models in utterances recognition and proficiency evaluation, which is quite different with the previous methods, using the same model in both steps. We call the proposed model as double-models pronunciation scoring algorithm model. There are some problems in traditional scoring methods. In standard algorithm in which normative model was adopted, because of the mismatch between nonnative speech and native model, recognition performance would be decreased greatly for Chinese students who are nonnative speakers, compared with native ones. In nonstandard solution in which nonnative model was adopted, pronunciation with low proficiency level may be rated as high quality one, as model was trained with imperfectly nonnative speech data. The presented method separates recognition from assessment stage. It can solve the problems that low recognition performance and score mismatch. The experimental results showed the proposed model reached a good outcome to evaluation English pronunciation for Chinese students.