Beyond Rhythm and Prosody: Inter-Utterance Silence Predicts Drone Movement Duration in Spontaneous Vocal Guidance

Abstract

Most existing voice-controlled robot systems in human-robot interaction (HRI) rely on fixed or predefined movement durations, or on explicit quantity markers in spoken commands, with no direct link between the spontaneous acoustic signal and the amount of movement produced. Prosodic features such as fundamental frequency, energy, and speech rate have been proposed as potential carriers of such information, but their validity in naturalistic, spontaneous speech remains largely untested. This paper investigates whether several acoustic and rhythmic parameters, including inter-utterance silence duration, are more strongly associated with drone movement duration than classical prosodic features. Using N = 1, 592 spoken instructions from spontaneous French speech collected during real human-drone teleoperation, we evaluate this relationship through correlation and regression analyses. Silence duration substantially outperforms all prosodic features, yielding the strongest correlation with movement duration (r =.498, p <.001) and accounting for approximately 80% of the variance explained by the full model (R2 =.226 univariate vs. R2 =.283 multivariate). These findings question the assumption that prosody encodes motor intent in continuous-control HRI, and suggest that when a speaker talks may matter more than how they sound.