Audio watermarking in DCT: Embedding strategy and algorithm

Where should watermarks be embedded in DCT domain in order for the audio watermark to be in inaudible and robust? Though many works in the literature agree that watermark should be embedded in the middle-frequency coefficients or perceptually significant components, how to determine the position embedded has no universal definition. A new embedding strategy and an algorithm applying the strategy are proposed for audio watermarking in DCT domain based on research of DCT noise model. By defining a noise sensitivity vector, a math relationship between the watermark embeddable position and the audible quality of the audio file with watermark is set up. First the optimal position to be embedded based on the inaudible requirement is selected. Second the power coefficient of watermark is adjusted to satisfy the requirement of robustness. So the utmost need of inaudible and robust of audio watermarking scheme was filled. It is a good approach to settle the trade-off between the two requirements of audio watermarking: inaudible and robustness. Experimental results demonstrate that the embedding strategy is effective and the watermarks thus generated are very robust.