Pronunciation modeling of loanwords for Korean ASR using phonological knowledge and syllable-based segmentation

Abstract

This paper aims to improve the performance of automatic pronunciation generation of foreign loanwords in Korean by using phonological knowledge and syllable-based segmentation. The loanword text corpus used for our experiment consists of 16.6K words extracted from the frequently used words in set-top box, music, and POI domains. At first, pronunciations of loanwords in Korean are obtained by manual transcriptions, which are used as target pronunciations. A syllable-based segmentation method considering phonological differences is proposed for loanword pronunciation modeling. Performance of the baseline and the proposed method are measured using PER/WER and F-score at various context spans. The result shows that the proposed method outperforms the baseline. We also observe performance decrease when training and test sets come from different domains, which implies that loanword pronunciations are influenced by data domains. It is noteworthy that pronunciation modeling for loanwords in Korean is enhanced by reflecting phonological knowledge. The loanword pronunciation modeling in Korean proposed in this paper can be used for (1) ASR of application interface such as navigation and set-top box and (2) computer-assisted pronunciation training for Korean learners of English.