Fill the silence! Basics for modeling hesitation

Abstract

In order to model hesitations for technical applications such as conversational speech synthesis, it is desirable to understand interactions between individual hesitation markers. In this study, we explore two markers that have been subject to many discussions: silences and fillers. While it is generally acknowledged that fillers occur in two distinct forms, um and uh, it is not agreed on whether these forms systematically influence the length of associated silences. This notion will be investigated on a small dataset of English spontaneous speech data, and the measure of distance between filler and silence will be introduced to the analyses. Results suggest that filler type influences associated silence duration systematically and that silences tend to gravitate towards fillers in utterances, exhibiting systematically lower duration when preceding them. These results provide valuable insights for improving existing hesitation models.

Disfluencies in Russian Spoken Monologues: A Distributional Analysis

“…Literature on these distributional issues is much more scarce, some few exceptions to be mentioned include e.g. [Crible et al 2017], [Betz et al 2015], [Betz and Kosmala 2019], [Bóna 2019]. Our paper is in line with this distributional approach, which we advocate and check against Russian data.…”

“…In examples, we provide the session ID (04, 22, or 23) and IDs of the EDUs involved (in these IDs, N stands for Narrator, C for Commentator, and R for Reteller). Since we were primarily interested in how fluent and disfluent stretches are distributed across relatively long speech intervals, we concentrated on monologic fragments and didn't consider dialogical parts 3 . Overall, we annotated 32 mins of audio that contained 4,780 words.…”

Rethinking (Dis)fluency Within the Scope of Interactional Linguistics and Gesture Studies

“…Let us now illustrate this model with an utterance taken from the SITAF Corpus (Horgues & Scheuer, 2015) which has been analyzed in detail in previous work on disfluency (Betz & Kosmala, 2019;Kosmala, 2021bKosmala, , 2021aKosmala et al, 2019).…”

Hesitations in Urdu/Hindi: Distribution and Properties of Fillers & Silences

“…Recent analyses have shown cross-linguistic differences in the duration of silences preceding and following fillers in an utterance. [18,17] found that there is an interplay between fillers and associated silences in English and German as silences following fillers are significantly longer compared with the preceding ones. In French, on the other hand, the silences preceding fillers have been found to be significantly longer [17].…”