A dynamic, feature-based approach to the interface between phonology and phonetics for speech modeling and recognition