Paw speech watermarking for compressed video

A method for embedding raw speech as watermark signal into a compressed sequence of images is proposed in this paper. The idea of using raw speech as a watermark to enhance the robustness of the embedded watermark arises from the fact that the raw speech contains a considerable amount of redundancies, and its contents can still be recognized after retrieval process, even if the watermarked data was badly attacked. As long as the raw speech, extracted from the attacked watermarked image frame, contains enough important information, its contents can still be intelligible. In the experiments, the raw speech was embedded into MPEG coded video stream, then the simulation of attacks was conducted. After the embedded speech was extracted, listening test was performed to evaluate the intelligibility contained in the extracted speech. The impression of the human perceptual system, as it tends to adjust and learn quickly to recognize the repeated speech, enhances the probability of recognizing the contents of the extracted raw speech precisely. The experimental results showed that our method successfully survives most common attacks.