Dimensional Design of Generative Emotive Sounds for Robots

Non-Linguistic Utterances (NLUs) are essential parts of emotive exchanges, not only in human-human interactions but also in the context of human-robot interactions. This research aims to deepen our understanding of generative emotive sounds for the domain of human-robot exchanges. We investigated the connections between certain audio qualities and the perception of emotional arousal and pleasure, designing a novel generative algorithm using musical and prosodic parameters capable of expressing a range of emotions and a dimensional model of emotion. To assess the design algorithm, we conducted an end-user evaluation in which participants were asked to interpret the emotive NLUs conveyed by robots. In the evaluation we examined 4 archetypal emotions: excitement, contentment, sadness, and anger. We placed participants’ responses within the pleasure-arousal affect grid to analyze the distinctness of the emotive sounds. The study revealed that participants consistently associated excited, sad and angry NLUs with significantly different emotional states but not for content NLUs. These findings contribute valuable insights into how to design generative NLUs which can enhance the emotional depth of human-robot interactions, with potential applications across various domains.