Automated Chinese Language Proficiency Scoring by utilizing Siamese Convolutional Neural Network and fusion based approach

Abstract

The previous approaches have failed to effectually score the language proficiency of a non-native speakers especially in case of non- English languages which are complex and a slight change of pronunciation can alter the nature of the word. In this study, we proposed an automated language scoring system to test the proficiency of Chinese language. We have employed a novel fusion approach of a 38-feature based model and a Siamese convolutional neural network (Siamese CNN) which can accuracy identify the difference between the native speech and the test taker's speech. The results show that out model have achieved comparable performance to the state of the art and solved the pronunciation problems as well. Furthermore, we have provided a fusion based approach and provided extensive amount of experiments which shows that our method is state of the art and can be utilized in real time Chinese language proficiency scoring.