I wonder why they don't compare it to their existing live voice model realtime-2.