SVD分解和离散小波域特征值量化的安全数字语音水印算法

Secure digital speech watermarking based on SVD decomposition and eigenvalues quantization in discrete wavelet domain

针对传统数字语言水印算法鲁棒性较差、复杂度较高等问题,提出一种基于SVD分解和离散小波域特征值量化的安全水印算法。该算法以离散小波变换的特征值量化为基础,利用离散小波变换将每帧数字语音转化到小波域,再利用SVD奇异值分解计算近似系数特征值,而不是细节系数部分;最后,使用量化后的特征值嵌入水印比特位信息。实验结果表明,当量化步长和所选帧长较高时,该算法能有效抵御过滤攻击、加性高斯白噪声攻击、重采样攻击和剪切攻击,其中高斯白噪声攻击和剪切攻击的误码率几乎为0。相比其他优秀算法,该算法具有更好的鲁棒性。

As the low robustness and high complexity problem of traditional digital speech watermarking algorithm, a digital speech wa- termarking based on SVD decomposition and eigenvalues quantization in discrete wavelet domain is proposed. Eigenvalues quantization in discrete wavelet domain is the base of the proposed algorithm, firstly, discrete wavelet transform is used to transform every frame of the digital voice into wavelet domain; Then, SVD singular value decomposition is used to calculate the eigenvalues of approximate coefficient, rather than part of the detail coefficients; Finally, the watermark bits were embedded by qnantization of the eigenvalue. The experimental results show that when quantization step size and the length of the selected frame are high, the algorithm can effectively resist filter at- tacks, additive white Gaussian noise attack, resampling and cropping attack, where the bit error rates of Gaussian white noise attacks and cut attacks are almost zero. Compared to other outstanding algorithms, the proposed algorithm has better robustness.