영어 강세 교정을 위한 주변 음 특징 차를 고려한 강조점 검출

Abstract

Prominence of speech, which is often called 'accent,' affects the fluency of speaking very much in American English. In this paper, we present an accurate prominence detection method that can be utilized for computer-aided language learning (CALL) system. We employed pitch movement, overall syllable energy, 300-2200 Hz band energy, syllable duration and spectral and temporal correlation as features to model the prominence of speech. After the features for vowel syllables of speech is extracted, prominent syllables are classified by SVM (Support Vector Machine). To further improve the accuracy, the differences in characteristics of neighboring syllables are added as additional features. We also applied a speech recognizer to extract more precise syllable boundaries. The performance of our prominence detector is measured based on Intonational Variation in English (IViE) speech corpus. We obtained 84.9% accuracy which is about 10% higher than previous research.