M-GFCC: Audio Copy-Move Forgery Detection Algorithm Based on Fused Features of MFCC and GFCC

Audio copy-move forgery has seriously affected the authenticity of audio, and copy-move forgery detection and localization of audio has become an urgent problem. In this paper, we propose an audio copy-move forgery detection algorithm based on the fusion of mel frequency cepstrum coefficient and gammatone frequency cepstrum coefficient feature fusion. Firstly, the algorithm extracts the voiced speech segments in the audio using the algorithm of voice activity detection, and then extracts the MFCC and GFCC features from each voiced speech segment, and then the corresponding weights are assigned to the two features for weighted summation to get the fused features. Finally, the similarity between the fused features of each voiced speech segment is calculated by using the dynamic time warping algorithm, and the part with the DTW distance less than the threshold is determined to be copy-move forgery. The experiments on publicly available replicated mobile forgery databases show that the algorithm not only enables precise localization of tampered audio, but also has high robustness.