![]() |
Modeling Pronunciation Variation for Automatic Speech RecognitionRolduc, The Netherlands |
![]() ![]() |
A popular heuristic technique for pronunciation modelling extends the lexicon with pronunciations for selected word phrases or multiword units in order to capture limited cross-word pronunciation effects without significantly increasing acoustic confusability or computational requirements. This paper describes methods for selecting a set of multiwords and for learning alternative pronunciations which are more representative of those found in fluent speech. The methods are evaluated on the Switchboard spontaneous speech task.
Bibliographic reference. Nock, H. J. / Young, S. J. (1998): "Detecting and correcting poor pronunciations for multiword units", In MPV-1998, 85-90.