中文摘要：

针对采用二值图像作为水印的数字水印算法所存在的安全缺陷,提出了一种基于语音特征生成半脆弱水印的内容认证算法。首先,将基音周期生成的半脆弱水印嵌入在每帧语音第1部分的离散小波变换（DWT）和离散余弦变换（DCT）构成的混合域;之后结合语音统计均值稳定特性及同步码技术,选取稳健的16位同步码作为同步标记,通过量化每帧语音第2部分的样本统计均值嵌入同步码。实验表明,该算法具有很好的不可感知性,对添加噪声、重量化等常规信号处理具有一定的鲁棒性,同时能够准确定位剪切、插入等恶意攻击的篡改位置。

英文摘要：

Aiming at the shortcomings of using binary images as watermarks,a novel semi-fragile watermarking algorithm for speech content authentication was proposed.The semi-fragile watermarks were generated from speech pitch periods,and were embedded into the discrete wavelet transform and the discrete cosine transform in the first part of each speech frame.Then,using statistics characteristics and synchronization codes,16 bits synchronization code was embedded to the statistical average value of speech-frames samples in the second part.Experimental results show that the semi-fragile watermarking scheme is not only inaudible,but also robust against additive noise,re-quantization and other common signal processing operations.At the same time,it can accurately locate the tampered regions which are caused by insertion attacks,deletion attacks and other malicious attacks.