Modeling the alignment of fundamental frequency contours

Abstract

Segmental factors can cause large temporal changes in local fundamental frequency (F0) movements associated with accented syllables (‘‘accent curves’’), but these changes are often not phonologically or perceptually significant. Yet, other factors can cause temporal changes that are smaller but nevertheless significant. A model is proposed according to which accent curves are (phonologically, perceptually) equivalent when they are generated from a common accent curve template using a common set of alignment rules. These alignment rules specify where the points making up the template are to be placed on the time axis, based on the durations of the segments in the syllable sequence with which the accent curve is associated. This model is shown to provide a detailed and accurate account of alignment of accent curves over a wide range of segmental configurations. The model is embedded in a superpositional framework, in which accent curves, segmental perturbation curves, and phrase curves are combined to account for complicated surface F0 contours. Height of accent curves is determined via a multiplicative model from factors related to prominence and position in the phrase.