MARK TATHAM and KATHERINE MORTON COMPUTATIONAL MODELLING OF SPEECH PRODUCTION: ENGLISH RHYTHM

In this paper we examine the treatment of English rhythm from both a theoretical and an experimental perspective. There are major shortcomings in the way not just rhythm but also prosodics in general is modelled; this is all too clear in various applications of the theory, particularly in computational areas such as speech synthesis (Keller and Keller 2002). Our objective is to begin characterising rhythm computationally within a hierarchically based model in which prosody is the framework for speech production in general. Such a model tightly integrates suprasegmental and segmental properties of speech in a well defined and well motivated binding process. Since the overall framework is the prosody the phonetic rendering of utterances depends on their prosodic structure; this includes the detail of the structure of the syllable – a unit of the prosody. The idea is not novel [see Firth (1948) on the prosodic structure, and Kahn (1976) and Gussenhoven (1986) on the structure of syllables, in particular the phenomenon of ambisyllabicity and detailed phonetic rendering], and it formed the basis of the conceptual design of our SPRUCE computational model (Lewis and Tatham 1991). The prosodic framework approach within SPRUCE is expressed in such a way that (a) it incorporates hooks for phonetic rendering with expressive content, and (b) it is transferable to a speech synthesis system for the kind of model testing associated with utterance synthesis. SPRUCE itself is n o t a piece of 'speech technology', rather it is a speech production model which, because of its c o m p u t a t i o n a l nature, lends itself to being the basis of a speech synthesiser proper.