First International Conference on Spoken Language Processing (ICSLP 90)

Kobe, Japan
November 18-22, 1990

Evidence for a Syllable-Based Model of Speech Timing

W. Nick Campbell

ATR Interpreting Telephony Research Laboratories, Kyoto, Japan

To test whether a measure of lengthening can be found which applies to segments uniformly within the syllable, instead of differentially with respect to segment type or position in the syllable, the durations of each segment in the SCRIBE 200-sentence phonetically balanced database were converted to standard normal form by subtracting the means and expressing the residuals in terms of their standard deviations to yield a mean of zero and a variance of 1 for each phoneme distribution. These normalised values were taken to represent the amount of lengthening or compression undergone by each segment relative to its elasticity. This paper explores the extent to which lengthening or compression measured in this way applies within the syllable, and shows that stress-induced lengthening may be differentiated from intonation-phrase-final lengthening, which is in turn different from the phonetically-motivated lengthening that is undergone by vowels preceeding voiced plosives.

Full Paper

Bibliographic reference.  Campbell, W. Nick (1990): "Evidence for a syllable-based model of speech timing", In ICSLP-1990, 9-12.