Fig. 2From: Supervised Attention Multi-Scale Temporal Convolutional Network for monaural speech enhancementProposed encoder-decoder network U\(^2\)-LSTM. Dashed lines indicate connected features in the channel dimensionBack to article page