Copy-paste Tamper Detection Algorithm for Audio Signals Based on Feature Spectrogram Fusion

Copy-and-paste tampering is characterized by manipulability and imperceptibility. Meanwhile, the existence of various post-processing operations makes audio tampering detection more difficult. For audio copy-paste tampering with post-processing, the paper proposed an audio copy-paste tampering detection algorithm based on the fusion of feature spectrograms. Firstly, the audio signal is converted into Hilbert spectrogram and Mel spectrogram respectively, and the two feature spectrogram images are weighted and fused to obtain a more robust feature spectrogram image. Further the keypoints are extracted from the fused feature image by utilizing Scale Invariant Feature Transform and matching the keypoints. Then the duplicate regions on the image are identified with agglomerative hierarchical clustering algorithm to implement audio copy-paste tampering detection. The experimental results show that the proposed method is highly robust to the various common post-processing attacks; Compared with previous state-of-the-art audio signal copy-paste tampering detection methods, the proposed method in this paper has higher detection precision, recall and robustness.