A new method of speaker localization using the filtered correlation

Speaker localization is defined as the determination of the coordinates of the speaker in relation to a point in space. It is achieved by using a comparison between the signals received by different microphones to estimate the position and eventually the identity of the speaker. In the present paper, we have conducted a multidisciplinary research project, which led us to several interesting results. For that purpose, we used the information given by two cardioids microphones, placed in opposition: one at the left and the other at the right of the speakers, in order to determine the position of the active speaker and try to supervise the audio-visual recording (eg. supervision of meeting-rooms). To achieve the speaker localization task, we have employed a new detection method, which we called: the filtered correlation method. The principle of this method is based on the calculation of the correlation between the two signals collected by the two microphones and a special filtering, which preserve only the pitch bandwidth. Experiments are done on several scenarios containing one, two or three speakers in the meeting-room. Results show an efficiency of the proposed localization method and a possible tracking by camera.