Watermarking Algorithm For Speech Content Authentication

As a method for digital multimedia protection, digital watermarking has been developed rapidly in recent years and it has been regarded as an effective solution to solve the security problem of multimedia. Digital watermarking is used for copyright protection, authentication of multimedia and so on. However, currently available speech watermarking technique is far beyond mature yet compared with the image watermarking. This thesis focuses on speech content authentication watermarking algorithm in order to distinguish various content preserving operations from malicious tampering. Experiments are performed to test the effectiveness of proposed algorithms.The main research work and contributions of this thesis are summarized as follows:(1) This thesis systematically describes the concept, framework, the technique requirements and other traits of fragile watermarking techniques, summarizes all kinds of existent audio watermarking algorithms, finally introduces the attacks and benchmarks of audio watermarking techniques.(2) A semi-fragile speech watermarking algorithm based on speech model is researched, which is realized by quantization of line spectrum pair (LSP) parameters. Experiments show that the algorithm is robust against amplitude scaling and semi-fragile to white noise addition, and thus suitable for speech authentication.(3) A speech content authentication watermarking algorithm based on SNR of speech codec in wavelet domain is proposed. The step of quantization is calculated by SNR so that the imperceptibility of watermark is guaranteed. Experiments show that the algorithm is robust against several operations, such as resampling, white noise addition, compression and so on, and is sensitive to malicious tampering.(4) A semi-fragile watermarking algorithm which takes speech feature as watermark is proposed. The decision is made by the difference between the feature of watermarked speech and extracted watermark from every speech frame. In order to enhance the robustness against content preserving operations, "multi-frame decision" is proposed while decision made by single frame is not work well.