Abstract
Temporal sequential dependencies in speech can be realized as negative correlations between the variations in duration of successive articulations; for example, if the consonant in a consonant‐vowel syllable is articulated overlong, the following vowel will often be overshort, making the syllable less variable in duration than the sum of the variabilities of its parts. Temporal dependencies in an articulatory interval may be measured by the interval's relative variance (RV), i.e., the variance of the interval's duration divided by its length. Under an hypothesis of no sequential dependency between articulations, RV is approximately constant for all intervals. As sequential dependencies increase, RV decreases; units of articulation then emerge as speech segments with low RV. RV also separates two kinds of dependencies. Since RV, unlike relative error, is sensitive to changes of scale, tempo variations in repetitions of an utterance will artificially inflate the RV of tempo‐related intervals (e.g., rhythmic phrases), leaving unchanged the RV of segments whose durations derive more from peripheral constraints (e.g., aspiration). The RVs of some longer intervals drop dramatically when tempo effects are removed, showing clearly the presence of rhythmic constraints in speech.

This publication has 0 references indexed in Scilit: