Parameters of unit-based measures of speech rate

Abstract

Speech rate is a useful concept for understanding variation in speech, but there are many different ways that it can be defined. One particularly common way of measuring speech rate involves counting the number of linguistic units that occur within a given time period. In these unit-based approaches, there are several parameters that may influence estimated rate, including the size of the window used to count units, the location of the window, whether the measure is expressed as an event rate or its reciprocal (average unit duration), as well as the type of unit that is counted, e.g. words, syllables, or phones. A corpus study was conducted to investigate how these and several other parameters influence the strengths of correlations between rate measures and target segment durations. It was found that phone-based rate measures tend to be more strongly correlated with segment durations than other measures, and that proper rates, defined as events per second, outperformed inverse rates, defined as average durations per event. In addition, it was found that including the intervals associated with target segments in the window used to calculate rate led to substantial artifactual increases in correlation. Overall the findings of the study can provide guidance on practices for measuring speech rate and bear upon theoretical issues involving the role of phonological structure in the control of speech production.