抗翻录攻击的鲁棒语音水印算法

Robust Speech Watermarking Algorithm Against Recapturing Attacks

水印为数字音频的版权保护提供了一种技术手段。然而,随着录用设备的普及,翻录攻击成为一种去除水印信息的有效方法。为了提高水印算法的安全性,提出了一种鲁棒的抗翻录攻击的数字语音水印算法。定义了离散余弦系数对数均值(discrete cosine transform coefficients logarithm mean,DCT-CLM)的特征,分析了该特征对翻录攻击的鲁棒性,并给出了基于该特征的水印嵌入方法。帧号和水印一起作为嵌入在各语音帧的信息,通过量化DCT-CLM方法将帧号和水印一起嵌入在各语音帧中。帧号用来同步各语音帧的内容,从同步的含水印语音帧中提取水印信息,从而进行溯源追踪。和常见的语音水印算法相比,该算法除了对去同步攻击的鲁棒性之外,还能够抵抗对敏感语音内容的翻录攻击。

Watermarking provides a technical mean for copyright protection of digital audio. However,with the popularity of recording equipment, recapturing attack has become an effective method to remove audio watermarks. In order to improve the security of the watermarking system, we propose a robust speech watermarking algorithm against recapturing attacks. Firstly, we define the discrete cosine transform coefficients logarithm mean(DCT-CLM) feature and get the conclusion that the changes of DCT-CLM feature are very small after recapturing attacks. Secondly, Frame number and watermark are embedded together in frames by quantifying the DCT-CLM feature. Frame number is used to resynchronize watermarked speech after the signal is subjected to de-synchronization attacks. If watermarked frame is synchronized, we extract watermark bits from the frame for resource tracing. Compared with other speech watermarking algorithms, the algorithm proposed in this paper is not only robust against de-synchronization attacks, but also robust against recapturing attacks.