A Two-Stage Mispronunciation Detection Approach for Computer-Assisted Pronunciation Training
In this paper, we propose a two-stage mispronunciation detection approach for computer-assisted pronunciation training. In the first stage, the selected phonological rules are used to cooperate with ASR to detect mispronunciations based on language transfer. Because the first stage detection can only deal with the pronunciation errors in the scope of the phonological rules, and detection performance is depressed with the imperfect phoneme acoustic model. The rescoring method based on duration normalized log posterior probability (NLPP) is employed in the second stage to identify the recognition speech unit again. Furthermore, a new F冄-score ranking criterion is proposed for the first stage to balance the mispronunciation coverage and recognition confusion, in the aim of minimizing the cost of total detection errors. The experiment shows that the method only with phonological rules gets a best performance of 19991 total detection errors, and the normalized log posterior probability method costs 22264 total errors. Finally, the two-stage detection approach can reduce the total errors to 19498.
Hua Yuan Junhong Zhao Jia Liu
Tsinghua National Laboratory for Information Science and Technology,Department of Electronic Enginee State Key Laboratory on Transducing Technology, Institute of Electronics, Chinese Academy of Science
国际会议
2011亚太信号与信息处理协会年度峰会(APSIPAASC 2011)
西安
英文
1-5
2011-10-18（万方平台首次上网日期，不代表论文的发表时间）