Context Aware Mispronunciation Detection for Mandarin Pronunciation Training

Mispronunciation detection is an important component in a computer-assisted language learning (CALL) system. Many CALL systems only provide pronunciation correctness as the single feedback, which is not very informative for language learners. This paper proposes a context aware multilayer framework for Mandarin mispronunciation detection. The proposed framework incorporates the context information in the detection process and providing phonetic, tonal and syllabic level feedback. In particular, the contribution of this work is twofold: 1) we propose to use a multilayer mispronunciation detection architecture to detect and provide mispronunciation feedback at the phonetic, tonal and syllabic levels. 2) we propose to incorporate the phonetic and tone context information in mispronunciation detection using vector space modelling. Our experiment results show that the proposed framework improves the mispronunciation detection performance in all three levels.