Abstract :
[en] This paper presents a contribution of phonetic tools focused on prosody for the study of stuttering. Two French-speaking persons who stutter (PWS) and one non stuttering person (NSP) have been recorded while performing a map task (explaining to a remote person, on the basis of a map, the way to go from a given city towards another) under varying auditory feedback conditions (Normal Audio Feedback - NAF - and Delayed audio Feedback - DAF - with 80 ms, 120 ms and 160 ms delays). The stuttering profiles of the two PWS are contrasted, since PWS1 is characterized by a large amount of repetitions (of phonemes, syllables, words and sentences) although PWS2 exhibits frequent blockings. The events we study are the acoustical phenomena corresponding to pronunciation units deriving either from the realization of syllables or from the production of isolated sounds or groups of sounds (syllables onsets, filled pauses etc.). They are classified into the categories of an ad hoc taxonomy. The ISI fluency index (based upon time differentiation of the syllabic intensity peaks) is applied to those phenomena. It reveals striking differences in the speakers' behaviours, depending upon the speaker profiles (NSP vs. PWS), the conditions they have to cope with while speaking (NAF vs. DAF) and the kind of stuttering. It has to be pointed out that the observed ISI distributions are far to be Gaussian; beyond the implications in terms of statistical procedures to avoid or favour, this suggests an underlying complex structure of causality, with specific groups of ISI values being related to specific phenomena (e.g., very short ISIs corresponding to clonic behaviour or very long ISIs associated with irrelevant or parallel discourse, etc.), presumably linked with varying psycholinguistic sub-processes and/or levels of language production.