Limits in FIR subband beamforming for spatially spread near-field speech sources