Fig. 4From: Sub-convolutional U-Net with transformer attention network for end-to-end single-channel speech enhancementArchitecture of AHA blockBack to article page