基于希尔伯特-黄变换的音频信号复制粘贴篡改检测算法

Copy-paste Tampering Detection Algorithm for Audio Signals Based on Hilbert-Huang Transform

复制粘贴篡改是音频篡改常用的篡改方式。当被篡改的音频段来自于同一音频且伴随着后处理操作时,这种篡改往往难以检测。针对有后处理的音频信号复制粘贴篡改,文章提出了一种基于希尔伯特-黄变换的音频信号复制粘贴篡改检测和定位的方法。采用音高跟踪技术将音频信号的有声段和无声段区分开来,对每个有声段分别实行希尔伯特-黄变换以提取谱图特征,进而利用动态时间规整(DTW)来计算每个谱图特征的相似度,当满足设定阈值条件时,通过比较希尔伯特谱图特征的DTW值来检测音频复制粘贴篡改,同时根据DTW值对应有声段的索引来定位复制粘贴篡改的位置。实验结果表明:该方法对几种常用的音频复制粘贴后处理操作都具有较好的鲁棒性;与基于梅尔谱图的音频信号复制粘贴篡改检测方法相比,所提方法检测准确率高、召回率高、鲁棒性好。

Copy-paste tampering is a common method for audio tampering,which is often difficult to detect when the tampered audio segment comes from the same audio accompanied by post-processing operations.For audio signal copy-paste tampering with post-processing,this paper proposes a method for detecting and locating audio signal copy-paste tampering based on the Hilbert-Huang transform.Firstly,the pitch tracking technique is used to distinguish the voiced and unvoiced segments of the audio signal, and the Hil-bert-Huang transform is applied to each voiced segment separately to extract the spectrogram features, and then dynamic time warping (DTW) is used to calculate the similarity of each spectrogram feature. When a certain threshold is met, the audio copy-paste tampering is detected by comparing the DTW values of the Hilbert spectrogram features and locating the location of the copy-paste tampering according to the index of the DTW value corresponding to the audible segment. The experimental results show that the method has good robustness to several common audio signal post-processing operations after copy-paste tampering. The proposed method has been demonstrated through extensive experiments to have higher detection accuracy, recall, and robustness compared with Kasi's proposed Mel-spectrogram-based audio signal copy-paste tam-pering detection method.