可恢复的数字语音取证水印技术研究

For the development of digital speech processing technology and the increasingly rich of the editing tools for digital speech signal, it is easy to attack and forge the digital speech. The existence of fake digital speech makes the digital voice faces the threat of losing trust, and has affected the credibility and public recognition of digital content expressed by digital speech signal seriously. Recoverable digital speech authentication technology based on digital watermark is an effective way to solve the problem. At present, the research of digital watermark used for speech forensics is focused on tamper location of hostile attack, while lack the research of tamper recovery for the attacked signal. The research target of the project is to recover the digital speech signal subjected to malicious attack, and the main research contents include: (1) Studying the security and precise tamper location speech authentication scheme based on watermarking; (2) Studying the generation method of the signal used to tamper recovery; (3) Presenting a high-capacity watermark embedding method for the signal used to tamper recovery, regarded as the watermark. This study will provide technical support for restoring the original meaning of the attacked digital speech and exploring the facts that attackers mean to cover up, and provide the useful reference for inferring the attacker’s intention and strong guarantee for constructing modern legalistic society. Meanwhile, this study can help to promote the digital speech authentication watermark technology popularization and application.

数字语音处理技术的发展和编辑工具的丰富，使对数字语音的篡改和伪造变得简单易行。伪造数字语音的存在使数字语音面临失去信任的威胁，严重影响了数字语音为载体的数字内容的可信度和公众认可度。可恢复的数字语音取证水印技术是解决该问题的有效途径。目前，数字语音取证水印的研究主要集中在对恶意攻击的篡改定位上，缺少对被攻击内容的篡改恢复。本项目以数字语音内容的篡改恢复为研究目标，主要研究内容包括：（1）安全的精确篡改定位数字语音取证水印方案研究；（2）用于篡改恢复的信号生成方法研究；（3）将用于篡改恢复的信号作为水印嵌入的大容量水印嵌入算法研究。这项研究为还原被攻击数字语音内容要表达的真实含义，挖掘攻击者有意掩盖的事实提供技术支持；为推断攻击者的意图提供有益参考，进而为构建法治社会提供有利保障。同时，也有助于推动数字语音取证水印技术的推广和应用。

数字语音处理技术的发展和编辑工具的丰富，使对数字语音的篡改和伪造变得简单易行。伪造数字语音的存在使数字语音面临失去信任的威胁，严重影响了数字语音为载体的数字内容的可信度和公众认可度。可恢复的数字语音取证水印技术是解决该问题的有效途径。目前，数字语音取证水印的研究主要集中在对恶意攻击的篡改定位上，缺少对被攻击内容的篡改恢复。本项目从数字语音真实性取证的实际需求出发，研究了数字语音的取证方案，并进一步研究了对被攻击语音内容的篡改恢复方法。经过三年的努力，课题组先后研究了安全的精确篡改定位数字语音取证水印方案；用于篡改恢复的信号生成，以及将用于篡改恢复的信号作为水印嵌入的大容量水印嵌入方案。提出了一系列可恢复的数字语音取证水印算法，实验验证了该技术有效的水印的取证能力以及对被攻击内容的篡改恢复能力。重新录音的方法可以去除含水印音频中嵌入的水印信息，可录音设备的普及使该问题更加凸显。针对这一问题，课题组进一步研究了抗翻录攻击的音频水印技术。.项目亮点如下：1、课题组在中国计算机学会推荐A类期刊IEEE Transactions on Information Forensics and Security发表学术论文一篇；2、课题组在国际权威期刊Signal Processing、Multimedia Tools and Applications等期刊发表学术论文三篇。