Speech Recognition and Intrinsic Variation (SRIV2006)
Foreign accented speech recognition systems have to deal with the acoustic realization of sounds by non-native speakers that does not always match with native models. Thus the standard native speech modelling alone is generally not adequate, and it is usually extended with models of phonemes estimated from speech data of foreign languages, and often complemented with extra pronunciation variants. Previous studies showed that such modelling techniques are effective. In this paper, several modelling approaches are briefly recalled and their performances in the recognition of non-native speech comprising multiple foreign accents, are analyzed. Afterwards, the adaptation of the acoustic models on various sets of non-native speech is studied. Results show that an adaptation on a limited set of foreign accents provides speech recognition performance improvements even on foreign accents absent from the adaptation data.
Bibliographic reference. Jouvet, Denis / Bartkova, Katarina (2006): "Analysis of model adaptation on non-native speech for multiple accent speech recognition", In SRIV-2006, 9-14.