基于时频域特征的直接扩频音频隐写分析

借鉴语音压缩编码的理论模型与方法，对wav音频预处理之后，在频域分析短时离散傅里叶变换（STFT）域的频谱响应曲线；在时域构造全极点模型，分析预测残差，得出了一组对音频直接扩频（DSSS）隐写高度敏感的时频域联合统计特征。并利用层次向量机（H-SVMs）设计二级分类器，提出了一种新的隐写分析方法，不仅可以检测音频是否进行了隐写，还能进一步检测隐写是在FFT、MCLT域，DCT、MDCT域还是DWT域进行的。实验结果表明，此方法对于多种变换域的数据隐藏方法，平均检测准确率达90％以上。

Steganalysis of DSSS Audio Based on Time-frequency Domain Characteristics

From voice compression coding theory models and based on the pretreatment of wav audio,the paper analyzes the spectral response curve of short-time Discrete Fourier Transform(STFT) in frequency domain;constructs all-pole model and analyzes the prediction residual in time domain.Thus a group of time-frequency domain characteristics highly sensitive to Direct Sequence Spread Spectrum(DSSS) audio steganalysis is achieved,and a 2-level SVM is designed according to the theory of Hierarchical SVMs(H-SVMs).Finally...