Sound source localization and signal separation for office robot "JiJo-2"

Abstract

A method of sound source localization and source separation for the sound sources in the near-field range is described in this paper. For the sound localization, the subspace method was employed to obtain a high resolution which is necessary for the range estimation with a small-sized array. The modified MVBF with the estimated localization information was used for signal separation. In an experiment conducted in a middle-sized meeting room, 60-70% of word recognition rate was achieved.