TDOA map adaptation in sound source localization

Abstract

Recently, many researchers have studied about adaptation methods of various environments since intelligent service robots generally have to interact with human beings in any environments. In this paper, an adaptation method of angle-TDOA map is proposed which can be applied to a sound localization system when a certain configuration error or deviation is brought up by the change of environments. In addition, two cases of situation are defined along with two suggested methods respectively to solve the problem. The first method is building an angle-TDOA map with a test signal generated at know positions when a microphone array is randomly placed. The second one is an on-line adaptation method for the real-time sound localization. In this method, we will use a vision system to find out the real direction of sound sources and the angle-TDOA map is learned from the difference between map and actual value of TDOAs. We use a face detection module to update an angle-TDOA map and assume the detected person is generating sounds. Finally experiments show the effect of the adaptation in both cases.