Abstract: Speech enhancement (SE) models based on deep neural networks (DNNs ... In this paper, a high-efficiency encoder-decoder structure, inspired by the top-down attention mechanism in human brain ...