Multipurpose speech watermarking based on multistage vector quantization of linear prediction coefficients

摘要

To make speech watermarking achieve both copyright protection and integrity verification, a novel multipurpose speech watermarking algorithm based on the multistage vector quantization (MSVQ) of linear prediction coefficients (LPCs) is presented in this article. The property of natural speech that the vector quantization (VQ) indices of the LPCs amongst neighboring frames tend to be very similar is utilized to embed the robust watermark in the indices of the first-stage VQ(VQ1). Then, the semi-fragile watermark is embedded in the indices of the second-stage VQ(VQ2) with index constrained VQ encoding scheme. Both the robust watermark and the semi-fragile watermark can be extracted without host speech. Simulation results verify the effectiveness of the proposed algorithm in terms of robustness and semi-fragility.