A Robust Blind Audio Watermarking Scheme Based on Singular Value Decomposition and Neural Networks

The paper proposes a novel blind audio watermarking scheme based on singular value decomposition (SVD) and neural networks (NN), which makes use of memorization and adaptation capabilities of NN. Experimental results show that the proposed watermarking algorithm is strongly robust against several kinds of attacks such as noise addition, re-sampling, re-quantization, MP3 compression, replace and so on. Additionally, the proposed algorithm achieves signal-to-noise radio (SNR) values ranging from 21dB to 48dB for different watermarked sounds and embedded strengths. These results demonstrate that the proposed watermarking method is a suitable candidate for audio copyright protection.