Automatic Scoring of Flat Tongue and Raised Tongue in Computer-assisted Mandarin Learning

In this paper, a paradigm for the automatic scoring of flat tongue and raised tongue in computer-assisted Mandarin learning is presented. In this paradigm, distinctive features of Mandarin is used to distinguish the two classes consonants and score them. The confusion of them mainly results from different articulation places. The method based on the distributing of concentrated frequency area is proposed for scoring these consonants. In this method, recursive calculation of searching maximum energy value with sliding window is used to remove the high energy area of low frequency and the ratio of average energy is used as feature to evaluate flat-tongues and raised-tongues. With the method, the correct rate of scoring for the two classes consonants can reach 98.35%.