Anti-Forensics of Double Compressed MP3 Audio

Introduction

With the popularization of portable recording equipment and the rapid development of multimedia technology, malicious tampering can be implemented to digital audio easier and easier. For example, it may obstruct the justice if the tampered audios are taken as evidences in court. Therefore, digital multimedia technology makes our lives convenient, and the application field of digital audio authenticity and integrity bring serious security issues. In addition, the content protection and integrity authentication of digital audio has become a hot research topic. Digital audio forensics (Jin, Wang, Yan, Ma, & Zhou, 2016) is to detect and identify the digital audio’s integrity by analyzing the statistical features of the detecting audio signal.

Various effective forensic methods have been proposed in recent years. Ikram and Malik (2010) proposed a method to determine the audio integrity based on correlation similarity measure. In Ly et al. (2013), the MFCC and logarithm MFCC were considered as the discriminative features to identify the recording environment. Malik and Zhoa (2012) and Hua et al. (2016) examined the integrity of audios based on the electric network frequency (ENF) signal existed in the given audios. As for the MP3 audio forensics methods presented in Yang et al. (2008) and Yang et al. (2012), it was found that frame structure of MP3 audios will be destroyed when the audios were tampered.

Since audio tampering would inevitably produce double compression. It has been attention to study for audio double compression. Yu et al. (2014) put forward a method of detecting fake-quality audios, which utilized the number of MDCT coefficient as the features. The results show that their scheme was effective for detecting double compressed audios. Liu et al. (2010) proposed to detect the fake-quality audio by using Huffman table index. The experimental results demonstrated that the method was effective. Yang et al. 2010) found that the MDCT coefficients whose values fall into (-1,1) of the double compressed audio were less than that of the single compressed audio, as well as the frame offset to detect whether the audio processed with double compression. The results show that the method has higher detection rate for the MP3 audio from low bit rate to high bit rate, but its performance became poorer for the cases of high bit rate to low bit rate and the same bit rates. Liu et al. (2010) integrated the number of non-zero MDCT coefficients, the average distance between non-zero MDCT coefficients and the distribution of zero and non-zero coefficients as a feature vector to distinguish the single compressed and double compressed audios. Compared with the high bit rate to low bit- rate, the low bit rate to high bit rate achieves higher detection accuracies.

The competition between forensics technology and its countermeasure anti-forensics (Chen, Xiang, & Huang, 2016; Milani, Piazza, & Bestagini, 2014) has escalated over the past few years. In the field of audio anti-forensics, the work (Chuang, Garg, & Wu, 2012; Chuang, Garg, & Wu, 2013) present the first anti-forensic scheme and its countermeasures for ENF-based audio authentication. Chuang et al. proposed various anti-forensic operations to resist tampering with the underlying ENF signal while preserve the quality of the host signal. Zhao and Malik (2013) and Zhao et al. (2016) put forward a method to remove audio carry environmental characteristic, which aims to cover the audio splicing tampered by environmental noise.