On The Exploration Of Active Forensics In Digital Audio

For the development of high-speed networks and the explosive increase of hardware using digital information, the users using digital information (digital images, digital audio/video and text, etc.) are increasingly growing at present, and distributed throughout every corner of the world almost. The digital information provides convenience for people’s life. Users can upload their own digital artwork (digital recording, digital photos, etc.), and download digital multimedia content they need, which enrich the people’s daily lives. However, the users worry about the wide range of threats to the digital information. For the popularity of personal computer and the increasingly rich of the multimedia editing tools, it is easy to edit, attack and forge the digital information. For digital multimedia, the attackers can almost attack the multimedia content in the absence of costs in order to achieve ulterior purpose, which can destroy the authenticity and integrity of the multimedia content, and the attack is imperceptible. Recent digital multimedia content forgery events reduced the digital multimedia information credibility. The research on the issue of digital information forensics has become a research focus in the area of multimedia information security.Considering the widely used of the digital speech in telephone communications, news stories, etc., and the actual demand of verifying the authenticity and integrity of digital speech signals, the exploration of active forensics in digital speech will be conducted. The detailes are as follows:1、Content-based digital watermark provide a method to solve the veracity and integrity of digital audio, which is vulnerable to feature-analyzed substitution attack. Based on pseudo-Zernike moments, a digital audio forensics algorithm robust against feature-analyzed substitution attack is proposed, in order to increase the security of the content-based digital watermark schemes. The watermark is generated by using the pseudo-Zernike moments of first segment for each frame, and embedded in the pseudo-Zernike moments by quantifying the sample values of the second segment. The algorithm has the advantages of that existed in content-based digital watermark schemes, increases the security, and has the ability of tamper localization.2、Coefficients self-correlation degree is a feature of digital speech signal, which reflects the degree of closeness of the coefficients. Experimental results indicate that the feature is high efficiency, and fragile to malicious attacks and robust to common signal processing operations. Based on speech coefficients self-correlation degree, a speech forensics algorithm is proposed, and the content-based method robust against desynchronization attacks is given. Comparing with the schemes robust against desynchronization attacks based on synchronous code, the algorithm reduces the load of watermark system and increases the security.3、Comparing with other orthogonal moments, such as Zernike moments, pseudo-Zernike moments, etc., Bessel-Fourier moments have many advantages, higher computational efficiency, more robust against signal processing. In order to make better use of the advantages and to solve the insecurity issue that the public features-based schemes are faced, the definition and fast computation of Bessel-Fourier moments of discrete signal are given, and the properties of Bessel-Fourier moments for discrete signal are analyzed. And based on Bessel-Fourier moments, a security speech forensics algorithm is proposed. Compared with the public features-based schemes, the security of the algorithm is improved. And the efficiency of the scheme proposed is higher than that based on Zernike moments and pseudo-Zernike moments.4、In order to increase the security of the encrypted data in the cloud storage, a forensics and recovery algorithm for encrypted speech in cloud computing is proposed. The definition of signal’s energy ratio is given, and the properties of it are analyzed. In the algorithm, the speech signal is scrambled and watermark is embedded in the scrambled signal before uploading onto the cloud storage. The users download the encrypted signal when needed, and authenticate the signal. If the contend is tampered, based on the tamper localization, substitution is performed for the attacked signal, and signals substituted are scattered over different segments after anti-scrambling transformation, and the meaning of the recovery signal is same to the original one in hearing. Theoretical analysis and experimental evaluation results show that the proposed algorithm is effective.