ISCA Archive - Modelling phones by microsegments in a phonetically oriented recognition system (original) (raw)

Modelling phones by microsegments in a phonetically oriented recognition system

B. Kaspar, K. Schuhmacher

A tentative system of subphone units, called 'microsegments', is presented. Microsegments are defined by articulatory features and are used to model the temporal structure of phones by simple pronunciation networks. The inventories of both phones and of microsegments are of moderate size but allow the description of common variants of pronunciation as well as the handling of coarticulation effects. The process of labelling speech according to these models requires more effort than labelling phonemes only, but remains reliable. Keywords: Phonetically oriented speech recognition, subword units, articulatory features, temporal modelling