Sound Source Localization Based on Audio-visual Information for Intelligent service Robots

Abstract

In this paper, we present an Sound Source Localization (SSL) based on audio-visual information with robot auditory system for a network-based intelligent service robot. The main goal of this paper is to combine audiovisual-based Human-Robot Interaction (HRl) components that can naturally interact between human and robot for SSL. The proposed approach includes two main steps. The first step performs speech recognition and sound localization to know whether the user calls the robot or not as well as the direction of the caller respectively, when someone calls robot's name. Here sound localization is based on GCC(Generalized Cross-Correlation)-PHAT(phase Transform) by frequency characteristics. In the second step, a robot moves forward to the caller based on face detection. The robot platform used in this work is wever-R2, which is a network-based intelligent service robot developed at Intelligent Robot Research Division in ETRI. The effectiveness of the proposed approach is compared with audio-based SSL itself.