Stress patterns, pauses, and timing in prose reading

A decomposition of a read text into a succession of interstress intervals reveals basic structures of stress timing and stress distribution within a sentence and is the bases for analysis of quasi-rhythmical phenomena, speech tempo, and reading style. In a case study, we describe how stress foot statistics may be adopted for the analysis of various reading modes such as normal, low voice effort, high voice effort, and distinct reading. INTRODUCTION The long term objective of our project is to l e k about what constitutes good reading style and to develop a perspective of the demands of a text and individual variations. We also intend to extend experience from prose reading to the study of poetry reading and relate our findings to the theory of musical composition and performance. In the course of our work we have found it necessary to devote a major part of our time to methodological problems combined with small scale pilot studies. As described by Fant, Nord, & Kruckenberg (1987); Fant, Kruckenberg, & Nord (1988); Fant & Kruckenberg (1986a, 1986b, 1988a, 1988b), we have developed quantitative continuous scales of syllabic stress, with reference to durations and FO patterns and also to subjectively scaled measures. A substantial amount of work has been devoted to studies of stress groups, i.e., interstress intervals or "stress feet". We have shown that such intervals, defined from the onset of a stressed vowel to the onset of the next stressed vowel, see Fig. 1, provide a convenient basis for analysis of rhythmical qualities and also of stressed-unstressed contrasts which is one of many parameters that enter descriptions of reading style. I I stressed vowels f interstress Fig. I . Interstress intervals are bounded by the onsets of stressed votr~els. interval STRESS GROUP ANALYSIS Interstress intervals are of the order of 500 ms, ranging from 0.25 sec to 1 sec with a standard deviation of about 35%. Thus, there is no absolute isochrony. However, according to Lehiste (1975), there still exists a trend of perceptual isochrony. The duration of a stress foot increases systematically with the number of phonemes or syllables contained. From a linear regression analysis of free feet, i.e., those that do not span a pause or an otherwise marked boundary, see Fig. 2, we may derive the foot interstress interval constants a and b, of which b is the increment of duration per additional phoneme and a is the sum of all stress induced segment lengthenings within the foot. Typical values are a=100 ms and b=59 ms per phoneme with an average foot duration of Ta=535 ms corresponding to 7.4 phonemes or 2.9 syllables. kc 1000 z STRESS INTERVALS 9 lTn = u t b a n < cr 000 3 n -1 -< 600 =I CK LL J 'Z Loo Tn=158rn.53 (r-0.96) /AJ. nine sentences n = number of phonemes CK o = meon volue of 1, IV1 2 4 6 8 10 12 14 16 NUMBER OF PHONEMES. n Fig. 2 . Duration of uninterrupted interstress intervals, 'pee feet", as a function of the number of phonemes contained. For our reference subject, we found a=158 ms and b=53 ms per phoneme and an average interstress interval of T a d 5 0 ms. Within a group of 15 subjects, we found a variation of a from 50 ms to 200 ms whilst b had a more restricted range of variation, from 50 to 65 ms per phoneme, which points at a general stability of the duration of unstressed phonemes and overall speech rate. The higher the a and the smaller the b, the greater is the contrast between stressed and unstressed syllables, which also conditions a somewhat lower relative spread of foot durations. An increase of a is associated with a greater overall increase of consonant durations in the stressed syllable than of the duration of the stressed vowel. Differences in inherent phoneme durations tend to average out within a stress foot. Besides the number of phonemes contained, the main source of variation of the duration of a stress foot is the particular degree of stress. We have quantified stress by a normalized syllable duration index which has a mean value of 2 for stressed syllables and 1 for unstressed syllables. The standard deviation of the duration index of syllables carrying the stress within the feet was 0.4. The essential variance of stress feet durations was the same as that of stressed syllables. Within a sentence, positive departures from the mean were compensated by negative deviations, see Fig. 3, and to a far greater extent than could be predicted from standard deviations of prediction errors. This tendency conforms with a notion of stress planning within the constraints of a total energy expenditure assigned to a sentence, 0hrnan (1976), and can also be interpreted as a tendency to minimize deviations from an overall synchrony plan.