A Model for Speech Unit Duration

Abstract
A version of the exponential decay process is advanced as a mathematical model for speech unit durations. Distributions of speech unit durations are reported from a variety of speech samples. It is argued that a simple exponential model provides a workable approximation to these distributions. A more exact model is also outlined which takes account of (1) the fact that extremely short speech units lasting only a small fraction of a second are relatively less likely to occur, and (2) the fact that there tend to be small shifts in expected speech unit duration within a particular speech sample.