数字语音拼接篡改检测技术研究

发布时间：2018-08-19 18:16

【摘要】：数字语音随着互联网的发展已经逐渐走进了人们的日常生活。人们不仅可以通过数字语音在网络上进行沟通交流,而且语音信息还作为人机交互的媒介正在迅速发展。数字语音作为常见的多媒体信息给我们生活带来极大的便捷,与此同时,音频编辑和处理软件的功能不断的完善使得对语音的编辑和修改变得更加容易和便捷,并且通常人耳难以分辨这些处理留下的痕迹,如果被非法利用或恶意传播将给社会稳定带来严重的威胁。因而,对语音信息的真实性、完整性验证变得尤为重要。本文对数字语音信息真实性和完整性验证中的关键问题进行了研究与探索,主要研究了数字语音双压缩检测的相关问题以及数字语音拼接篡改检测技术。具体的研究工作可分为以下四个方面:1.数字语音的篡改往往会导致双压缩过程,并且双压缩的码率与单次压缩相同。因此,本文首先研究数字语音相同码率的双压缩检测问题,利用MP3编码过程中的量化频谱系数作为特征,分析相邻两次压缩之间的变化特点,实现对双压缩的检测。2.数字语音的双压缩检测也会带来反取证问题,也即使得双压缩的语音被检测为一次压缩。本文通过研究语音压缩遗留的痕迹,提出一种通过添加零值采样点的简便方法移除了压缩的痕迹,从而实现了双压缩的反取证研究。通过对该问题的研究为后续篡改检测提供了有力支撑。3.数字语音的拼接篡改检测是本课题的研究重点,通过对语音的双压缩检测的研究,发现篡改操作同样会移除压缩痕迹,因此本文利用篡改前后压缩次数的不一致性实现了篡改定位,实验结果表明具有一定效果,但定位的准确性受定位的精度影响较大。4.基于压缩历史不一致性的篡改定位方法在定位准确性和定位精度方面都不是太高,为进一步提高定位精度和准确性。本文还提出利用编码的量化特性进行篡改定位,将零值频谱系数作为特征分析其量化前后的变化特征,提取了一种有效的特征实现了对篡改的检测,实验结果表明检测准确率达到了99%。
[Abstract]:With the development of the Internet, digital voice has gradually come into people's daily life. Not only can people communicate and communicate on the network through digital voice, but also voice information is developing rapidly as a medium of human-computer interaction. Digital speech, as a common multimedia information, brings us great convenience in our life. At the same time, with the continuous improvement of the functions of audio editing and processing software, it is easier and more convenient to edit and modify the voice. And the human ear is usually difficult to distinguish the traces of these treatments, if used illegally or spread maliciously, it will bring a serious threat to social stability. Therefore, for the authenticity of voice information, integrity verification becomes particularly important. In this paper, the key problems in the verification of the authenticity and integrity of digital speech information are studied and explored, and the related problems of digital speech dual compression detection and digital speech mosaic tampering detection technology are mainly studied. Specific research work can be divided into the following four aspects: 1. Digital speech tampering often leads to double compression, and the bit rate of double compression is the same as that of single compression. Therefore, this paper first studies the problem of double compression detection of digital speech at the same bit rate. Using the quantization spectrum coefficient in the process of MP3 coding as the feature, the variation between adjacent compression is analyzed, and the detection of double compression. 2. Double compression detection of digital speech also brings the problem of anti-forensics, that is, the dual compression speech is detected as one compression. In this paper, by studying the traces left by speech compression, a simple method of removing the traces of compression by adding zero sampling points is put forward, which realizes the research of anti-forensics with double compression. Through the study of this problem, it provides a strong support for the subsequent tamper detection. 3. The detection of digital speech splicing tampering is the focus of this research. Through the research of the double compression detection of speech, it is found that the tampering operation will also remove the compression trace. Therefore, the tamper localization is realized by using the inconsistency of compression times before and after tampering. The experimental results show that it has some effect, but the accuracy of the location is greatly affected by the accuracy of the location. The tamper localization method based on compression history inconsistency is not too high in the accuracy of location and positioning, in order to further improve the accuracy and accuracy of location. This paper also proposes to use the quantization characteristic of coding to locate the tamper. The zero-value spectrum coefficient is used as the feature to analyze the change features before and after quantization, and an effective feature is extracted to detect the tampering. The experimental results show that the detection accuracy is 99%.
【学位授予单位】：宁波大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN912.3

【参考文献】