Research And Realization Of Audio Authenticity Identification Methods

With the rapid advance in digital audio technology, we are able to easily capture digital audio signals, but also can use audio editing software to easily edit and modify them. If the intentional or unintentional tampered audio data is used in forensic or other important occasions, it will likely lead to certain undesirable social issues, so researching into authenticity identification of digital audio data is particularly important.This paper primarily focuses on research of digital audio authenticity identification technology. Two types of identification methods to respectively detect audio copy pasting and up-sampling forgery are given. The main work is as follows:(1) For digital audio copy-paste operation in the same audio, this paper presents an identification method based on Gammatone filter bank and SURF feature. Firstly, the method calculates the digital audio sonogram based on Gammatone filter bank and extracts all SURF feature points from it. Then, the feature points set is randomly divided into two sub-sets of feature points. KD-Tree searching algorithm is used to match feature points between two sub-sets of feature points, and for each sub-set of feature points, recursion is operated until all possible matching feature points pairs are sought out. Finally, gives a Main-Horizontal-Offset method to remove the pseudo matching pairs. The digital audio is identified as a copy-pasting-suffered one or not depending on whether some matching pairs are found, and the copy-pasting clips of audio can be located according to their distribution area of matching pairs.(2) For digital audio up-sampling operation, this paper presents an identification method based on changes of sub-band spectrum. Counting the characteristic that up-sample can’t increase the frequency information, while the high frequency distribution is less and more smooth, this method firstly puts pre-emphasis on audio, and then identifies audio up-sampling tamper detection by analyzing distribution of the sub-band frequency spectrum and figures out the original sampling rate based on the ratio of the number of sub-band with lager spectrum amplitude to the total number of sub-band.Experiment results verify the effectiveness of two audio authenticity identification methods presented in this paper.