Digital Audio Forgery Detection Methods Based On Correlativity And Spectrum Analysis

Due to the development of digital audio techniques and the popularization of professional audio processing software, it tends to be that people can obtain digital audio materials more conveniently and can edit or revise digital audio files according to various purposes. However, editions or modifications have unnegligible impact on the integrity, dependability and authenticity of those digital audio files. Especially where the authenticity of the file is strongly needed, distorted or even forged audio may cause unpredictable harm. Therefore, improving and perfecting digital audio detection technology plays a significant role in verifying the credibility of digital audio data.This paper focuses on passive detection of digital audio technology and carries out research on the copy and paste operation of the same segment of the audio. WAV format audio files being as the main research object, detection methods from three aspects are stated as follows:(1) Methods of appraising digital audio based on matching degree analysis. Of this type, some research and implementations have been carried out on the method based on zero-mean normalized cross-correlation and the method based on Pearson correlation coefficient is proposed as well. By calculating the matching degree of random two words of the audio under detection and comparing the results with the proper thresholds, these two methods could identify the copy and paste operation. Meanwhile, by conducting simulation experiments, the accurateness and robustness of the raised methods are tested, and detection conclusions under different experimental conditions are proposed as well.(2) Methods of appraising digital audio based on spectrogram analysis. Of this type, the proposed two kinds of detection methods based on spectrogram analysis solve the copy and paste problem from the aspects of phase correlation and Euclid Distance. These two methods divide the audio and visualize the signal by drawing the spectrogram and extracting the frequency spectrum information on different time quantum, calculate the phase correlation degree between random two words of the drawings respectively according to phase correlation and Euclid Distance and identifies the falsify parts which are through copy and paste operation.(3) Methods of appraising digital audio based on frequency spectrum analysis. In this part, the detection method based on higher order statistics analysis is proposed, which uses discrete wavelet transform to extract the feature of each syllable, calculates the matching degree of the Pearson coefficients between random two syllables, makes higher-order statistics analysis on the falsify parts where the matching degree is greater than the set threshold, and finally identifies the copy-and-copied relation of the fields. Meanwhile, the detection method based on electric network frequency is introduced in this chapter. This method uses the property that when the original audio is distorted, the incidental electric network frequency is changed as well, to implement phase estimation of the incidental electric network signal of the audio, and to estimate whether the audio under detection is distorted according to the phase diagram of the signal.The computer simulation experimental results show that the proposed methods can detect copy and paste field precisely, and possesses robustness on general degrading operations.