Modeling and automatic detection of English sentence stress for computer-assisted English prosody learning system

概要

論文の詳細を見る
As prosodic features play an important role in human communication, learning prosodic patterns is essential to acquire English pronunciation. We present a Computer-Assisted English Prosody Learning System for Japanese students. Pronunciation evaluation is done by automatic detection of sentence stressed syllables which compose stress-timing rhythm. Syllable HMMs are categorized based on error patterns of the stress. The modeling makes it possible to generate effective error diagnosis and instruction. We also propose a method of multi-stage discrimination mat reflects native speakers' perception as weights of acoustic features. The method achieves a stress detection rate of 95.1% and 84.1% for native speakers of American English and those of Japanese, respectively.
社団法人日本音響学会の論文