Segmental timing control in speech production

Abstract: This paper presents a quasi-formal model describing the accuracy with which the durations of individual segments are controlled in speech production. The model assumes, in accordance with previous data, that when a speaker intends a fixed speech rate he actually speeds up and slows down by small but measurable amounts over a period of minutes. A little-known variance statistic, the half mean square successive difference, is employed to separate the component of variability due to these slow changes in rate away from variability due to errors in segmental duration control; the accuracy of the speaker’s segmental clock is then shown to be characterized by a constant of proportionality, termed “relative variance,” between segment variability and segment duration. The model is then tested, with fair success, against the measured durations of various segments in two sentences repeated many times by nine normal English speakers. Two possible models for the neural mechanism underlying the segmental clock are discussed, namely a “capacitor discharge” model and a “clock and counter” model, and possible differences between them are explored.