In order to construct a CALL system that can teach learners accent and intonation of Japanese, automatic identification of accent types and intonation types for learner's sentence utterances was investigated. With an aim of realizing a visual feedback perceptually meaningful to learners, a course investigation was conducted on the acoustic (prosodic) features of speech taking their effects on human perception into account. As for the accent type identification method, the use of target values of F0 in mora final was evaluated. The method was proved to be with good correspondence to human perception of high/low pitch. As for the intonation identification, 6 types of sentence final intonation were selected, each conveying different type of intention, and the relationship between sentence-final acoustic features and perceptual impression was investigated.
展开▼