A Mispronunciation Detection Method of Confusing Vowel Pair for Chinese Students

Abstract

This paper presents a discriminative features based mispronunciation detection method for confusing vowel pair /i/ vs /I/ that are frequently mispronounced by Chinese learners of English. Firstly, the mean of the 39-dimensional Mel Frequency Cepstral Coefficients (MFCC) feature vector over all the frames of the current phoneme segment is employed as features to characterize the phoneme. Secondly, many specific acoustic features that can effectively capture the crucial properties of the long and short vowels are extracted. Finally, the Support Vector Machine (SVM) classifier is used for discrimination between confusing vowels /i/ and /I/ by using the discriminative features extracted from each phoneme. The experimental results show that, the proposed method can produce higher accuracy than the traditional Automatic Speech Recognition (ASR) based methods. In addition, the combination of spectral features with specific acoustic features can achieve better performance than using individual features.