Decision of the timing of turn-taking based on the prediction of the end of older adult’s speech during scenario-based dialogue with a voice-calling robot

Abstract

Toward a voice-calling robot system that helps older adults maintain their physical and cognitive functions and quality of life, we develop a scenario-based dialogue system which is intended to detect their health and living conditions based on their responses, including speech to the robot. Turn-taking is important to keep the conversation last longer with this type of system. To allow the robot system to conduct a turn-taking that a user feel easy to speak to the robot, we have proposed the system which learns the individual differences in the silent section This paper reports a method for predicting the end of the speech of a user by learning individual differences in the silent section during a speech of the user, and a dialogue experiment with the method. In the experiment, some users felt that they were able to have a dialogue at the right time, while others felt that it was difficult to adjust the timing. In the future, we will study a system that takes into consideration the interaction with the user.