基于音频特征MFCC的零水印算法  

ZERO-WATERMARKING ALGORITHM BASED ON AUDIO FEATURES OF MFCC

作　　者：宋慧[1] 李晨[1] 田丽华[1] 

机构地区：[1]西安交通大学软件学院,陕西西安710049

出　　处：《计算机应用与软件》2016年第10期140-144,共5页Computer Applications and Software

基　　金：国家自然科学基金项目(61403302)

摘　　要：利用小波域的低频部分的奇异值具有比较稳定的性能,设计一种以音频特征为基础的零水印方案。该方案利用原始音频特征——Mel频率倒谱系数(MFCC),一方面来构造水印,另一方面还被用于音频内容认证。先将原始音频做分段操作,取各分段进行DWT变换;选取小波域的低频部分,并做SVD操作;最后利用奇异值平均值实现水印的"嵌入",获得一个位置索引向量并据此提取水印信息。仿真实验显示,该方案在完全不影响音频听觉质量的前提下,在抵抗常见音频攻击时有很强的鲁棒性,并能进行恶意篡改定位。By using the stability of singular value of low frequency coefficients on wavelet domain,we designed a zero-watermarking scheme which takes the features of audio as the basis. The scheme uses the feature of the original audio—mel frequency cepstrum coefficients( MFCC) to construct the watermark sequence,in addition this feature is also applied to audio content authentication. In the algorithm,original audio is divided into sections at first,and each section is carried out the discrete wavelet transform. Then the low frequency parts of wavelet transform are selected for singular value decomposition. Finally the mean of singular value is used to implement the ' embedment' of watermarks in original audio,and this obtains a position index vector,it can be used to extract the watermark information. Simulation experiment indicated that the scheme presented in this paper has quite strong robustness on resisting some popular attacks and can locate the malicious tampering on the premise of without losing auditory quality.

关 键 词：MFCC 零水印 奇异值分解 篡改定位 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]