The Concept and Measurement of Speech Rhythm

Abstract

This chapter discusses the nature and measurement of speech rhythm and proposes a multidimensional model of speech rhythm. Acoustic research on speech rhythm usually relies on so-called rhythm metrics, and the existing metrics are described and compared. Most rhythm metrics are based on duration and quantify, for example, the variability of vocalic durations. Greater variability is associated with stress timing and less variability with syllable timing. However, in order to fully grasp the rhythm of a language or variety, other acoustic correlates of prominence should also be taken into account to quantify the various dimensions along which speech rhythm can vary. These include variability in intensity, loudness, fundamental frequency, and the variability of sonorant and voiced durations. More generally, a syllable-timed language can be characterised as consisting of recurrent elements (such as syllables or vocalic intervals) of relatively equal prominence. Usually not all correlates of prominence contribute equally to such a syllable-timed rhythm. A language might be more syllable-timed with regard to one correlate, for example, the variability of vocalic durations, but more stress-timed with regard to another, for example, variability in intensity. It follows that speech rhythm can only be captured adequately by a multidimensional model that acknowledges the possibility of different coexisting rhythms in a language. The chapter also makes suggestions to improve methods for the quantification of speech rhythm. First of all, additional rhythm metrics are developed that are based on the variability of sonorant and obstruent intervals, based on the conclusion that the acoustic contrast between sonorants and obstruents is much more salient than the contrast between vowels and consonants. Furthermore, the chapter argues that apart from variability in intensity, variability in loudness also needs to be considered. In a next step, co-occurrence patterns of different correlates of prominence need to be taken into account. This point is illustrated by developing a rhythm metric that quantifies the simultaneous variability of duration and loudness. Broadly speaking, two kinds of languages/varieties are conceivable. In one, prominent vowels might be long and loud at the same time, so that duration and loudness reinforce each other as correlates of prominence. In another language or variety, prominent vowels might be either long or loud, so that duration and loudness partially offset each other in prominence. The potentially reinforcing use of duration and loudness as correlates of prominence can be accounted for by devising a rhythm metric that quantifies their simultaneous variability.