Abstract: Speech enhancement models based on deep learning, such as the complex U-Net model, have achieved good results. However, traditional methods based on convolutional neural networks often ...