即时语音通信下高效语音感知哈希认证方法研究

发布时间：2018-08-16 18:06

【摘要】：随着网络通信技术、移动通信技术和手机通信技术的快速发展,语音由其采集的方便、自然和技术的成熟,成为人们即时通信的主要方式之一。然而在语音即时通信过程中,语音易受到编解码、信道噪声、延迟、以及丢包的影响,并且由于无线网络信道的开放性,为黑客窃听、恶意篡改等非法攻击打开了方便之门。为了实现高效的语音认证,并且解决语音感知哈希在鲁棒性、区分性、认证效率之间的相互影响,以及实现语音的篡改定位,论文主要研究了即时语音通信下的高效语音感知哈希认证算法,以及在即时通信高速及低可靠的条件下,实现具有降噪、篡改检测与定位的高效语音感知哈希算法。主要研究内容如下:1.针对即时语音通信对感知哈希算法的高效率要求,这就需要感知哈希算法在具有良好的内容保持操作鲁棒性和不同语音区分性的前提下,使得算法具有认证数据量小,运行效率高。为此,提出了一种基于线性预测最小均方差(linear prediction-minimum mean squared error,LP-MMSE)的高效语音感知哈希认证算法。该算法在对内容保持操作有良好的感知鲁棒性和不同语音的区分性的同时,还具有摘要性小、运行效率高等优点。另外,为了解决内容保持操作的感知鲁棒性、不同语音的区分性和认证效率之间的平衡问题,提出了一种基于改进谱熵法的高效语音感知哈希认证算法。通过将Teager能量算子分别结合线性预测系数(linear prediction coefficient,LPC)、LP-MMSE和线谱对(linear spectrum pair,LSP)系数等方法的实验对比,该算法在鲁棒性、区分性和认证效率等方面具有良好的平衡性,且感知哈希摘要性小、能够满足语音通信实时认证的要求。2.针对目前异构移动终端语音通信格式相互转换后对语音内容保持操作的感知鲁棒性和不同语音的区分性,以及实现降噪、小范围篡改定位和高效认证性等要求,提出了一种基于谱减与能熵比法的高效语音感知哈希认证算法。该算法首先对待认证语音进行预处理,再进行谱减法降噪;然后对降噪之后的语音进行能熵比法计算得到每帧的能熵值参数矩阵;最后进行哈希构造生成二进制感知哈希序列。实验结果表明,该算法有效地降低了语音的噪声,对语音内容保持操作具有良好的感知鲁棒性和不同语音的区分性,且感知哈希的摘要性小,认证效率高、能够实现小范围的篡改检测与定位。
[Abstract]:With the rapid development of network communication technology, mobile communication technology and mobile phone communication technology, the convenience of voice acquisition and the maturity of nature and technology have become one of the main ways of instant communication. However, in the process of voice instant communication, speech is vulnerable to the influence of coding and decoding, channel noise, delay and packet loss. Due to the openness of wireless network channel, it opens the door to illegal attacks such as hacking and malicious tampering. In order to achieve efficient speech authentication, and to solve the interaction between robustness, discrimination, authentication efficiency and speech tampering localization, speech perceptual hashing is solved. This paper mainly studies the efficient speech perceptual hashing authentication algorithm for instant speech communication, and implements an efficient speech perceptual hash algorithm with noise reduction, tampering detection and location under the condition of high speed and low reliability of instant communication. The main research contents are as follows: 1. In view of the high efficiency requirement of the perceptual hash algorithm in the instant speech communication, it is necessary for the perceptual hash algorithm to have a small amount of authentication data under the premise of good content keeping operation robustness and different speech discrimination. High operational efficiency. For this reason, an efficient speech perceptual hashing authentication algorithm based on linear predictive minimum mean square error (linear prediction-minimum mean squared) LP-MMSE is proposed. This algorithm not only has good perceptual robustness and different speech distinctiveness to content keeping operation, but also has the advantages of low summarization and high efficiency. In addition, in order to solve the problem of perceptual robustness of content keeping operation, the balance between different speech discrimination and authentication efficiency, an efficient speech perceptual hash authentication algorithm based on improved spectral entropy method is proposed. By comparing the Teager energy operator with the linear prediction coefficient (LP-MMSE) and the linear spectrum method with the (linear spectrum pairn LSP coefficient, the proposed algorithm has good balance in robustness, discrimination and authentication efficiency, and has less perceptual hashing summarization. Can meet the voice communication real-time authentication requirements. 2. In view of the perceptual robustness of voice content retention operation and the distinction of different speech, the requirements of noise reduction, small range tampering and location, and high efficiency authentication are also presented in view of the current voice communication formats of heterogeneous mobile terminals. An efficient speech perceptual hash authentication algorithm based on spectral subtraction and energy entropy ratio is proposed. The algorithm preprocesses the authenticated speech first, then reduces the noise by spectral subtraction; then calculates the energy entropy parameter matrix of each frame by using the energy entropy ratio method after the noise reduction; finally generates binary perceptual hash sequence by hash construction. The experimental results show that the proposed algorithm can effectively reduce the noise of speech, have good perceptual robustness and distinguish different speech, and have less summarization of perceptual hashing and high authentication efficiency. Can achieve small range of tamper detection and location.
【学位授予单位】：兰州理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN912.3

【参考文献】