From syntax to acoustic duration: A dynamical model of speech rhythm production

This paper presents a speech rhythm production model able to generate segmental acoustic duration from several levels of dynamical coupling between linguistic and production-related subsystems. A probabilistic algorithm for phrase stress assignment accounts for both prominence and constituency prosodic relations by considering the coupling between a dependency-grammar system of markers and constituent-size constraints. This algorithm copes with intra- and inter-speaker prosodic variability. Having as input the position and magnitude of underlying phrase stress, and a set of dynamical control parameters, the model acts at three nested temporal domains to assign segmental duration in Brazilian Portuguese. The modelled V-to-V duration patterns reproduce the patterns found at the surface under several conditions of perturbation. The nature and advantages of the dynamical model of speech rhythm production for simulating natural data are thoroughly discussed.