Besides, due to the unavailability of manually labeling the fine-grained sequential correspondence between audio-text pairs, we attempt to model ATR as a cooperative game process to flexibly handle ...