The effect of rehearsal on filled pause production in short presentation speech

Spontaneous conversations and presentations often contain filled pauses like um and uh, yet their functional role and acoustic realization remain poorly understood. We investigated the effect of rehearsal on filled-pause production. Participants delivered a 5-minute slide-based presentation comparing coffee and tea consumption and repeated the task three times in succession. Speech recordings were automatically transcribed and force-aligned for acoustic analysis. Results show that rehearsal did not lead to a uniform reduction of filled pauses. Instead, hesitation behavior was redistributed across grammatical embedding sites: mid-phrase hesitations decreased substantially, while fillers increasingly clustered at phrase and predicate onsets, indicating a shift from local execution difficulties to higher-level structural planning. This redistribution was accompanied by a differential reduction of filler types, with uh occurring less frequently overall and disappearing more rapidly across rehearsal rounds than um. Acoustically, filled pauses became significantly shorter, lower in intensity, and exhibited higher spectral tilt. Their mean F0 remained stable across rounds, but F0 variability and range decreased significantly. At the level of the full presentations, rehearsal led to fewer and shorter silent pauses and an increased speech rate. Thus, rehearsal does not simply eliminate filled pauses but reshapes their distribution and acoustic realization, resulting in increased fluency.