Perceptual evaluation of pronunciation quality for computer assisted language learning

Abstract

In this paper, we propose a novel method of perceptual evaluation of pronunciation quality for Computer Assisted Language Learning used in e-learning. The overall score of the pronunciation quality is the combination of the matching score, the perceptual score and the asymmetric score. The matching score is the measure of the acoustic distortion of the test speech, the perceptual score models the perceived distortion by human in perception domain and the asymmetric score describes the asymmetric effect of the sensation of the deletion error and the insertion error in spoken English. The correlation coefficient between the predicted objective score and the subjective score by the experts is 0.75, which is advantageous over current methods based on HMM. © Springer-Verlag Berlin Heidelberg 2006.