Detection of copy move forgeries in audio recordings using passive verification techniques

With the increasing number and ease of use of editing tools, attacks on audio files have become quite common, especially by malicious users. Considering that such audio files can be used in various environments today, both as evidence in forensic environments and for visibility on social media, proving their authenticity is of great importance. Many audio verification methods have been proposed in the literature to determine digital audio authenticity. In general, these methods are evaluated in two main categories: active and passive verification methods. In passive methods, audio verification is performed using statistical features extracted from the audio, and no additional information is required as in active methods. This study, based on passive verification methods developed within the scope of the thesis, aims to detect copy-paste forgeries in audio files. The audio file taken from the input is segmented using the proposed speech activity detection method. Feature extraction was performed from the obtained segments using PLP, Rasta-PLP, MFCC, and Spectral methods. Correlation, Euclidean distance, and DTW were used in the similarity calculation between the extracted features. The proposed forgery localization algorithm also marked the forged segments in the suspicious audio file. Experimental results obtained on two separate forgery databases produced from the TIMIT and Arabic Speech Corpus datasets show that the method is quite high compared to the studies in the literature. In addition, the performance of the proposed method was also tested with audio files attacked in the forgery databases. The results obtained show that the proposed method is quite robust against attacks applied by attackers to cover up forgery traces.