A schematic diagram illustrating the dual-branch architecture. The upper module CRN-Time employs a CRN to predict the temporal representation, the lower module CRN-SRS employs the same structure to predict the frequency-domain representation. After each layer of convolution, a bridge layer is used for feature conversion and fusion. Each module get an output and the output of CRN-SRS represents the outcome of the proposed network because of its better performance in the evaluation stage.