Use of high-level linguistic constraints for constructing feature-based phonological model in speech recognition

Modeling phonological units of speech is a critical issue in speech recognition. In this paper, we report our recent development of an overlapping feature-based phonological model which gives long-span contextual dependency. We extend our earlier work by incorporating high-level linguistic constaints in automatic construction of the feature overlapping patterns. The main linguistic information explored includes morpheme, syllable, syllable constituent categories and word stress markers. We describe a consistent computational framework developed for the construction of the feature-based model, and discuss use of the model as the HMM state topology for speech recognizers.