Fig. 4From: Speech enhancement by LSTM-based noise suppression followed by CNN-based speech restorationBlock diagram of the speech restoration network utilizing a CED network including encoder and decoder sub-networks with strided and transposed convolutions (tr-setup), where /2 indicates a stride of 2. The sizes of the feature representations inside the network regarding the frequency, frame, and feature map axis, respectively, are shown at the input and output of each layerBack to article page