15th International Congress of Phonetic Sciences (ICPhS-15)
Recently interest in rhythm has revived and a number of new empirical measures have been proposed, for example by Low and Grabe, and Ramus. Closer examination shows these models to be incomplete, however, with respect to both empirical and formal adequacy conditions. The present contribution addresses these issues in relation to hierarchical temporal structuring and proposes an integrated computational phonetic approach, introducing an empirical data mining heuristic for inducing rhythm timing trees from large quantities of time-annotated data. New algorithms are proposed for Timing Tree Induction (TTI), and for a Tree Similarity Index (TSI) to estimate the similarity between syntax parse trees and prosodic trees as predictors for the structure of the temporal TTI trees. A preliminary quantitative evaluation shows a preference for tail-heavy (iambic) tree branching in a read-aloud narrative. Applications in speech genre analysis and for duration modelling in speech synthesis are envisaged.
Bibliographic reference. Gibbon, Dafydd (2003): "Computational modelling of rhythm as alternation, iteration and hierarchy", In ICPhS-15, 2489-2492.