Invertible networks-based method with dual-embedding and attack layer for robust audio watermarking

Digital audio watermarking aims to embed information into audio and accurately extract it from the watermarked audio. Traditional methods rely on algorithms designed based on expert experience to embed watermarks into the time domain or transform domain of signals. With the development of deep neural networks, neural audio watermarking based on deep learning has emerged. Compared with traditional algorithms, neural audio watermarking achieves better robustness by considering various attacks during training. However, current neural audio watermarking methods suffer from low capacity and poor imperceptibility. Furthermore, the issue of watermark locating, which is extremely important and even more pronounced in neural audio watermarking, has not been adequately studied. In this paper, we design a dual-embedding watermarking model based on an invertible neural network for efficient localization. We also consider the impact of the attack layer on the invertible neural network in robustness training, improving the model to enhance both its reasonableness and stability. Experiments show that the proposed model can withstand various attacks with higher capacity and more efficient locating ability compared with existing methods.