Sixth International Conference on Spoken Language Processing
As non-native speakers become more frequent users of speech recognition applications, increasing the tolerance of the system with respect to non-native pronunciation and language use is important and is currently the focus of research in a variety of contexts. Dictionary modification, acoustic model adaptation, and acoustic model manipulation are a few of the techniques that have been reported successful in improving recognition of non-native speech. In this paper, we address the specific case of Japanese-accented English, describing the lexical and acoustic modeling techniques that give the best recognizer performance. We find that automatically generated pronunciation variants perform as well as hand-coded "golden" variants in reducing recognizer error, and that a significant improvement in system performance can be achieved with acoustic models retrained on a small amount of accented data.
Bibliographic reference. Tomokiyo, Laura Mayfield (2000): "Lexical and acoustic modeling of non-native speech in LVSCR", In ICSLP-2000, vol.4, 346-349.