[3N1-GS-8a] Robot and real worlds

9:30 AM - 9:45 AM JST(12:30 AM - 12:45 AM UTC)

[3N1-GS-8a-03] Robot behavior generation that conveys the atmosphere of real conversations

〇Sora Kiyoto¹, Atsushi Nakazawa¹ (1. Graduate School of Interdisciplinary Science and Engineering in Health Systems)

Keywords:

Robot,conversation,Large Language Models,Prosodic Control,Gesture Generation

Reproducing the “atmosphere” of human conversation in Human–Robot/Agent Interaction (HRI/HAI) is an important challenge. In this study, we define conversational atmosphere as “the overall impression that an observer receives from the entire conversation,” and hypothesize that it is formed through the integrated control of utterance content, vocal prosody, and gestures. Based on this hypothesis, we annotated multimodal data of human manzai-style dialogues and designed a generation script that integrally controls prosodic parameters, gesture motions, and their timing according to the utterance content. Using the proposed system, dialogues were reproduced by two social robots. As a result, compared to a condition in which only utterance content and timing were reproduced, the proposed method gave the impression of more closely replicating the atmosphere of real conversations. This study demonstrates the effectiveness of multimodal integrated control for reproducing conversational atmosphere in HRI.