A multi-modal approach for determining speaker location and focus