Estimating Timing of Head Movements Based on the Volume and Pitch of Speech

Abstract

Our research aims to create two friendly communication robots to talk with elderly people in nursing facilities. If the robots synchronize their head movements in response to the elderly person, the elderly person may react favorably to the robot. Then, the elderly person can enjoy talking with these two robots. In this paper, we investigated whether the volume and pitch of the speech are useful data for estimating the timing of head movements. Because the robots need to move their heads in real time, when one of the robots or the person is talking, we focus on the volume and pitch of the speech, not the content. Moreover, it was cleared which machine learning method creates suitable classifier models for estimating the timing of head movements. The experimental results showed that Random Forest classifier was the most suitable method.