Improving Speech Recognition on a Mobile Robot Platform through the use of Top-Down Visual Queues

Robert J. Ross, R. P. S. O'Donoghue, Gregory M. P. O'Hare

Abstract: In many real-world environments, Automatic Speech Recognition (ASR) technologies fail to provide adequate performance for applications such as human robot dialog. Despite substantial evidence that speech recognition in humans is performed in a top-down as well as bottom-up manner, ASR systems typically fail to capitalize on this, instead relying on a purely statistical, bottom up methodology. In this paper we advocate the use of a knowledge based approach to improving ASR in domains such as mobile robotics. A simple implementation is presented, which uses the visual recognition of objects in a robot's environment to increase the probability that words and sentences related to these objects will be recognized.