Multi-source Sound Localization using the Competitive K-means Clustering

Sound source localization is an important part of intelligent robot auditory system. It makes a robot to respond naturally to human user's call. In the ordinary situations, there always exist multiple sound sources including user's call. Since localized outputs from each source are mixed in distribution, clustering is an important issue in the multi-source sound localization. In this work, we propose a new k-means clustering algorithm for unknown number of clusters, which is the competitive k-means. We compared its performance to the adaptive k-means++ algorithm and verified its effectiveness. Finally, we applied it to our sound source localization for multi-source sound localization and achieved satisfying results.