基于相位谱的翻录语音攻击检测算法

发布时间：2018-01-12 08:29

本文关键词：基于相位谱的翻录语音攻击检测算法　出处：《电信科学》2017年08期 　论文类型：期刊论文

【摘要】：因与原始语音具有高度相似性,经高保真设备回放的翻录语音常被不法分子用于对说话人认证(ASV)系统进行攻击,以达到非法认证的目的。为提高系统抵抗翻录语音攻击的顽健性,通过研究原始语音与翻录语音产生的实际过程,发现两者在频率域相位上有明显差异,并在此基础上提出了一种基于相位谱的翻录语音检测方法。分析讨论了FFT和不同偷录、回放设备对翻录语音检测率的影响。实验结果表明,该方法能够准确地判断待测语音是否为翻录语音,其检测率达到了99.04%。并且,将该算法加载到说话人识别系统中,使系统的等错误概率(EER)降低了约22%,有效提高了系统抵抗翻录语音攻击的性能。
[Abstract]:Because of the high similarity with the original speech, the replay voice replayed by the high fidelity device is often used by the illegal elements to attack the speaker authentication / ASV system. In order to achieve the purpose of illegal authentication, in order to improve the robustness of the system to resist the voice attack, it is found that there are obvious differences in the frequency domain phase between the original speech and the recorded speech by studying the actual process of producing the original speech and the ripping speech. Based on this, a phasing-based speech detection method is proposed, and the effects of FFT, different recording and playback equipment on the detection rate are discussed. The experimental results show that. The method can accurately judge whether the speech under test is a copyrighted speech, and its detection rate reaches 99.040.Furthermore, the algorithm is loaded into the speaker recognition system. The equal error probability of the system is reduced by about 22%, and the performance of the system is improved effectively.
【作者单位】：宁波大学信息科学与工程学院;
【基金】：国家自然科学基金资助项目(No.61672302,No.61300055) 浙江省自然科学基金资助项目(No.LZ15F020010,No.Y17F020051) 宁波大学科研基金资助项目(No.XKXL1405,No.XKXL1420,No.XKXL1509,No.XKXL1503);宁波大学王宽诚幸福基金资助项目~~
【分类号】：TN912.3
【正文快照】： 1引言在信息时代,录音设备(如录音笔、智能手机等)日趋低价和高保真,回放设备也越来越精准,为录音造假提供了可能。翻录语音经录音设备偷录、回放设备回放,与原始语音难辨真假,一些说话人认证系统也无法辨别。而且翻录语音因偷录设备体积小、易偷录、成功率高等优势,已成为攻

【相似文献】