Abstract: It is challenging to deploy Transformer-based audio classification models on common terminal devices in real situations due to their high computational costs, increasing the importance of ...