Fig. 3From: Training audio transformers for cover song identificationThe training accuracy with or without average poolingBack to article page