Accent Type Recognition of Japanese Using Perceived Mora Pitch Values and Its Use for Pronunciation Training System

Through perceptual experiments, the fundamental frequency (F0) in mora unit was defined which well corresponded to the perceived pitch value. Several candidates, given as the combination of periods of observation and methods of F0 calculation, were tested. Based on the definition, a method was developed for the accurate recognition of Japanese lexical accents, and was applied to a system for teaching nonJapanese learners pronunciation of Japanese accents. In the method, each accent type was represented as a multidimensional Gaussian model, where F0 change between two adjacent morae was used as the feature parameter. The system first recognizes accent types of words in a learner's utterance, and then notices the learner if his/her pronunciation is good or not with audio and visual corrective feedbacks. Using TD-PSOLA technique, the learner's utterance is corrected in its prosodic features by referring to teacher's features, and offered to the learner. Since the learner can hear how his/her utterance should sound after correction, he/she can obtain a better idea on the correction as compared to the case where only teacher’s utterance is offered as the audio feedback. The visual feedback is also offered to enhance the modifications that occurred. Accent type pronunciation training experiments were conducted for 8 non-Japanese speakers, and the results showed that the training process could be facilitated by the feedbacks especially when they were asked to pronounce sentences.