In the field of robust audio watermarking, how to seek a good trade-off between robustness and imperceptibility is challenging. The existing studies use the same embedding parameter for each part of the audio signal, which ignores that different parts may have different requirements for embedding parameters. In this work, the constraints on imperceptibility are first analysed. Then, we present a segment multi-objective optimization model of the scaling parameter under the constrained Signal-to-noise ratio (SNR) in Spread spectrum (SS) audio watermarking. Additionally, we adopt the Nondominated sorting genetic algorithm II (NSGA-II) to solve the proposed model. Finally, we compare our algorithm (called SS-SNR-NSGA-II) with the existing methods. The experimental results show that the proposed SS-SNR-NSGA-II not only provides flexible choices for different application demands but also achieves more and better trade-offs between imperceptibility and robustness.

No abstract