Audio Zero-Watermarking Algorithm based on Beat Tracking

In order to protect the copyright of digital products effectively, digital watermarking technology is proposed. At present, synchronization attack is still a challenging problem in audio watermarking technology. In this paper, a robust zero watermarking scheme is proposed. Firstly, the beat activation function is obtained based on the bidirectional Long Short-Term Memory(BLSTM) recurrent neural network, and the beats of music are extracted by peak detection. Then the audio is segmented according to beat points. After framing for each segmentation, the mean value of each frame's normalized sampling values is calculated to obtain the mean value sequence. By encoding, the mean value sequence is converted into a binary sequence as the audio feature, which is xor with the watermarking image to generate the final zero-watermarking. Experiments show that the proposed scheme can effectively resist some synchronization attacks, such as TSM and jittering.