Modeling Those F-Conditions - Or Not

After several disappointing preliminary attempts to make condition-speci c models, we decided that it would be more advantageous to spend our time just trying to improve the core recognition system and use general adaptation techniques to deal with variations. This simpli ed the system immensely and freed up people and also made the transition from the PE system to the UE system much easier. We describe our attempts at condition-speci c modeling/adaptation/training. We show that the bene t for channel-speci c modeling is smaller than that for general adaption procedures and we argue that the cost is too high.