Computer-assisted English prosody learning system based on automatic detection of sentence stress and stress-timing

摘要

Computer-Assisted English Prosody Learning System for Japanese learners is presented. Learners' pronunciation is evaluated by automatic detection of sentence stressed syllables and foot durations which compose stress-timing rhythm. Syllable HMMs are categorized based on error patterns of stress. This modeling makes it possible to give effective error diagnosis instruction. And we propose a method of multi-stage discrimination that reflects native speakers' perception. As the result of experiences, the rate of stress detection have reached 95.1% and 84.1% for native and Japanese speakers, respectively. Furthermore, foot templates are constructed from native speech for stress-timing evaluation. Foot templates are categorized based on speaking rate, the number of syllables, syllable structure and so on. The timing score by templates are compared with the human judgement of English teacher.