Selective attention for voice matching and recognition