Ustubioglu, Beste; Küçükuğurlu, Büşranur; Ulutas, Guzin

Copy-move forgery is one of the most widely used methods in the field of audio forensic because it is difficult to detect and convenient to apply. However, post-processing operations applied to forged speech to hide traces of forgery make detection of forensic more difficult. In this study, we propose a robust method for detection and localization of the audio copy-move forgery using modified discrete cosine transform (MDCT). For this purpose, we first divide the speech recording into voiced and unvoiced parts by extracting a pitch sequence from the speech recording. After the determination of voiced parts, we extract MDCT coefficients from them and take the mean of the transpose of the coefficient matrix as the feature. These MDCT features are very robust against commonly used post-processing operations (especially audio compressing). Euclidean distance (ED) is applied to compute the similarities between the features. The voiced parts, which give minimum ED are determined as copy-move audio parts. Also, two separate databases (Pitch-based and Voice Activity Detection (VAD-based) are created during the work to evaluate the performance of the proposed method because there is no common database in the field of audio forensic in the literature. Experiment results show that the proposed method gives better results for detection and localization of audio copy-move forgery on different databases compared to other studies in the literature. The proposed method is also robust against common post-processing operations such as noise addition, filtering operation, and especially compression operation.

Kucukugurlu, Busranur; Ustubioglu, Beste; Ulutas, Guzin; Herencsar, N.

Due to the advantages of convenience, and difficulty of detection, copy-move forgery is one of the most common audio forgery forms. It is significant to decide whether there is a duplicated segment in the audio. In order to detect copy-move forgery, a new method is proposed in this paper. In the proposed algorithm, first the audio is segmented into syllables using a pitch tracking method. Second, each syllable analyzed with a 1-D local binary pattern operator. This operator calculates the histogram of each syllable. Mean square error is used to compute the similarities of histograms. The syllables which have similar histograms are detected as forged audio segments. Compared to another 1-D local binary pattern method, the proposed method has better detection results for copy-move forgery. Our experiments show that the proposed algorithm is feasible and effective, and robust against many common postprocessing operations such as noise addition, filtering, and compression.