Study On The Techniques Of Digital Audio Forgery Detection

With the rapid development of digital audio technology, there are more and more widespread applications in all aspects of social life. But the truth and integrity of digital audio are broken by the wide use of kinds of digital audio edit softwares. When the forged audio recordings are applied to some special occasions, it will have serious impact on the stability of social life. So digital audio forgery authentication needs to develop quickly and be improved.This paper focuses on the digital audio authenticity identification technology, and it proposes three types of detection methods. The detailed content is as follows:(1) Directing at the interpolation of digital audio, this paper proposes a detection method based on Singular Value Decomposition. This scheme points out that the linear correlation among the sample points of digital audio after interpolation is stronger than before. So depending on this character, the method carries out Singular Value Decomposition of the signal and counts the number of zero singular value in the singular value vectors to quantify the linear correlation. Finally, according to the comparison with threshold, it can detect the forgeries.(2) Directing at the copy and paste forgery of digital audio, two detection methods are proposed. The first method utilizes Zero-Mean Normalized Cross Correlation to quantify the distance between two words and decide the occurrence of forgery or not. The second scheme firstly use the dimension reduction character of Discrete Wavelet Transform and Singular Value Decomposition to obtain the feature of every word. Then it compares the Euclidean distance among the features with the special threshold and draws a conclusion that there exists forgeries or not.(3) Directing at the insertion or deletion of digital audio, this paper presents a detection method based on Electric Network Frequency analysis. Electric network signal is embedded into audio when recording. When audio suffers forgeries, the electric network signal must suffer the same kind of forgery at the same time and its phase is sure to have a sudden change at the forgery point. This paper estimates the phase of the electric network signal, draws up the plot of phase and locates the forgeries through the sudden changes of the plot of phase.Experimental results demonstrate the validity of the three forgery detection methods.