Adversarial Signal Denoising with encoder-decoder networks

ABSTRACT In this work, we treat the task of signal denoising as distribution alignmentbetween the clean and noisy signal. An adversarial encoder-decodernetwork is proposed for denoising signals, represented by a sequence of measurements. We rely on the signal's latent representation, given by the encoder, to detect clean and noisy samples. Unlike the standard GAN training, we propose a new formulation that suits to one-dimensional signal denoising. In the evaluation, we show better performance than the related approaches, such as autoencoders, wavenetdenoiser and recurrent neural networks, demonstrating the benefits of our approach in different signal and noise types. Index Terms-- signal denoising, adversarial learning 1. INTRODUCTION In signal processing, the presence of noise is a common problem independentfrom the signal type. One way to recover the signal is to use neural networks for denoising.