Mispronunciation detection based on cross-language phonological comparisons

This paper presents a method using speech recognition with linguistic constraints to detect the mispronunciations made by Cantonese learners of English. The predicted pronunciation errors have been derived from cross-language phonological comparisons, which are used to generate the erroneous pronunciation variations in a lexicon. The acoustic models are trained with native speakerspsila speech and used for recognizing the phone sequences, given the orthographic transcriptions. The experiments have examined that the agreement between automatic mispronunciation detection and human judges is over 84% for 21 Cantonese speakers.

1. Introduction

This work is investigating the automatic mispronunciation detection method to effectively highlight pronunciation errors made by Cantonese (L1) learners of American English (L2). The aim is to provide interactivity with regard to English teaching and self-learning in a computer-assisted language learning (CALL) system. In China, it would be especially useful to develop a CALL system with remedial instructions, since a large number of Chinese learners have no chance to talk with the native speakers to practice and then correct their pronunciations. However, most previous studies on CALL systems were designed to give a pronunciation measurement for non-native speakers [1], [7], [3], [9]. These methods incorporating with speech recognition techniques focus on accessing non-native speakers' pronunciation quality in a good or poor level. Other studies, such as [5], were conducted to discriminate a confusing pair of phonemes, e.g. the correct English pronunciation and its pronunciation marked by non-native speaker's accent.