This project aims to examine existing, commercial off-the-shelf (COTS) software for pronunciation training, identify key areas of pronunciation improvement for Chinese learners of English, as well as develop new technologies that can effectively support pronunciation training.
Description of process and deliverables
We surveyed over 20 selected COTS software and invited suggestions their users. Most systems support perceptual training but few support productive training. Provision of interactive instructions or corrective feedback for the learner is generally lacking. We have designed and collected a speech corpus named CU-CHLOE (Chinese University Chinese Learners of English). It consists of approximately 350 hours of English speech recordings from native Cantonese and Mandarin speakers (100 in each case). Practicing language educators were invited to examine the data.
Evaluation of outcomes
The analyses, together with systematic phonological comparisons between the primary (i.e. Cantonese / Mandarin) or secondary (i.e. English) languages, enable us to identify key areas of pronunciation improvements at the phonetic, phonemic and lexical levels. These are explicitly modeled in a home-grown automatic speech recognizer (ASR) named CHELSEA. This system can perform phonetic recognition on the learner’s speech input, detect phonetic mispronunciations and highlight erroneous areas.
Dissemination of results & deliverables
Our work has been disseminated through 7 published conference papers, a forthcoming journal paper, a keynote lecture (at IEEE ICALIP 2008), several invited talks at CUHK and Waseda University and numerous technology demonstrations (e.g. China Hi-Tech Fair).