Echo Cancellation Based on Improved Wave-U-Net

Most current implementations of deep learning-based echo cancellation algorithms use the short-time Fourier transform domain, but this approach may be limited by frame size selection and lack of accurate phase information. To overcome these limitations, in this paper, we will use the Wave-U-Net temporal network model as the baseline model for single-channel echo cancellation. Wave-U-Net performs better than traditional methods and neural networks in speech processing tasks; however, in the echo cancellation domain, the model still underperforms in near-end microphone speech feature extraction, and its ability to low distinction between near-end and far-end speech features, which is easy to cause confusion. Based on this, this paper will optimize the structure of Wave-U-Net network, aiming to improve its efficiency and performance of echo cancellation.