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

This article is a general idea of a statistical model for speech recognition given where phonetic and phonological knowledge sources are drawn from the current understanding of the global characteristics of human speech communication. They are effortlessly incorporated into the structure of a stochastic model of speech. A steady statistical formalism is presented in which the associated models for the discrete, feature-based phonological process and the continuous, dynamic phonetic process in human speech production are calculations at a crossing point. Two primary ways of executing the speech model and recognizer are presented, one based on the trended hidden Markov model (HMM) or explicitly defined trajectory model, and the other on the state-space or recursively defined trajectory model. Both executions build into their respective recognition and model-training, production affiliated trajectories across feature defined phonological units. The continuity and the constraint structure in the dynamic speech model permit a joint characterization of the contextual and speaking style variations manifested in speech sounds, in so doing holding promises to overcome some key limitations of the current speech recognition technology.

A dynamic, feature-based approach to the interface between phonology and phonetics for speech modeling and recognition. vol 24. ISS 4, Li Deng