This paper describes a system to identify the direction of user's keyword utterance for autonomous mobile robot. The robot is activated by the user's keyword utterance and identifies the speaker by face recognition. To capture the speaker's face within the camera's viewangle, the robot first have to identify the approximate direction of the utterance using acoustic information. To achieve the above-mentioned task, this system identifies the direction of keyword utterance by multiple microphone within the range of 45 degree. As this system is built into the mobile robot, hardware requirement is very tight due to battery restriction and space factor restriction. Therefore we developed the system which doesn't need expensive calculation. The system was evaluated by recall and precision using several thresholds. From the experimental results it is found that the length of the keyword dominates the absolute threshold value. Using mora-by-mora threshold, more than 80 recall and precision was obtained.
展开▼