Learned Dictionaries-Based Watermarking for Speech Authentication

Robust watermarking have been widely investigated for solving the problems of speech integrity authentication. In this paper, we introduce a novel robust watermarking algorithm for speech integrity authentication. In the watermarking algorithm, for embedding robustness, the approximation coefficients for stationary wavelet transforms are modified by patchwork and quantization index modulation, the robustness is further obtained through learning the approximation coefficients under learned dictionaries. The efficiency of the watermarking algorithm is shown via several experimental results. The good inaudible is first shown by signal-to-noise ratio and objective difference grades. Then, the robustness is shown by against various common signal processing operations whereas keeping inaudibility. The proposed scheme is sensitive to malicious modification of speeches, and simultaneously tolerate some common signal processing. Compared with state-of-the-art algorithms, the proposed algorithm obtains a better comprehensive performance in robustness.