Towards an interactive pronunciation dictionary for learners

초록

Research towards developing an interactive pronunciation dictionary that features sensitivity to learners’ native phonology is presented for the specific case of Korean learners of English as a Foreign Language (EFL). A speech corpus was collected to determine whether native-language (L1) dependent feedback on pronunciation improved speech quality as indicated by human perceptual tests. The feedback messages address segmental and prosodic problems in the speech of Korean EFL learners. The segmental phonetic variations that were deviated from native English phonological rules were analyzed to trigger the feedback messages on insertion, deletion and substitution problems of phones. The SUMMIT speech recognizer’s capability to model phonetic variations through explicit, context-sensitive, phonological rules was leveraged to detect pronunciation problems common to Korean EFL learners. Prosodic features of the speech were quantified to determine suitability for detection of stress, rhythm, and intonation problems. Taken together, the segmental and prosodic error detection methods were substantiated for generating specific feedback messages to improve pronunciation quality.

키워드