Computer assisted spoken English learning for Chinese in Taiwan

Abstract

The paper proposes an approach to computer assisted spoken English learning for Mandarin Chinese speaking people in Taiwan. Various studies have suggested the importance of acoustic models for pronunciation assessment. For English and Chinese people, their mother tongues are different; therefore, the corresponding spoken English, as well as the acoustic models, are also different due to subtle differences in pronunciation. Our aim is to have robust acoustic models and better phoneme segmentation in the recognition phase of the assessment. The proposed approach improves the speech recognition rate, leading to better reliability of HMM log-probability and the higher accuracy of phoneme segmentation. These two factors, in turn, contribute to the success of our pronunciation assessment system, as demonstrated in the experimental results.