基于录制环境检测的数字音频取证研究

Digital audio forensics based on recording environment detection

为解决数字音频伪造和篡改的检测问题,针对数字音频取证中录制环境检测,提出一种基于梅尔倒谱系数(MFCC)结合小波包分析的特征提取算法。算法用以提取音频的频域统计特性,结合时域特征构造特征集合,运用基于期望最大化(EM)的机器训练方法对音频录制地点进行分类和判断,实现数字音频录制环境的取证。实验结果表明,该算法能够较好的区分不同环境下的音频特性,纯净分类(无其他环境下的音频混入聚类组)最高可达98%。

To solve the detection problem of the forged or tampered digital audio, a feature extraction algorithm based on MEL cepstrum coefficients （MFCC） combined with wavelet packet analysis is presented according to digital audio forensic recording environment detection. Frequency domain statistical features are extracted by the algorithm mentioned above combined with the time domain features to structure a feature set. And then a machine training method based on expectation maximization （EM） algorithm is applied to detect the recording environment of the audio. A series of experimental analyses and tests show that the algorithm can distinguish the audio characteristics in different environments, and the best result is 98% with no audio recorded in other environments.