A single channel speech enhancement algorithm based on VGG19-RAFFUNet-HESE and FED-former model

VGG19-RAFFUNet-HESE cannot fully obtain the long-term information in the time/frequency domain conversion. Therefore, a single channel speech enhancement algorithm combining VGG19-RAFFUNet-HESE model and FED-former model is proposed. Firstly, FED-former model is embedded in the encoder and decoder of the original model. The Fourier enhancement module of FED-former model and the seasonal trend decomposition method are used to capture the global information of time series better in the coding layer. Then the depth and depth features are fused in the jumping connection part. Finally, the decoder part realizes the long sequence prediction in the forward transmission of information. The experimental results show that the algorithm can fully extract the information of long time series in frequency domain, and reduce the error of predicting long time series and the degree of speech distortion.