虚拟场景中篡改语音检测方法的研究  

The Study of the Tamper with the Speech Detection Method of the Virtual Scene

作　　者：吴为胜[1] 张良[2] 李磊[1] 

机构地区：[1]萍乡学院网络与教育技术中心,江西萍乡337055 [2]黄河科技学院信息工程学院,郑州450061

出　　处：《计算机仿真》2016年第7期418-422,共5页Computer Simulation

基　　金：2015年江西省教育厅科技项目(GJJ151264);2014年江西省教学改革研究课题(JXJG-14-22-9);郑州市科技局科技发展计划项目(20140663);郑州市嵌入式系统应用技术重点实验室建设项目(121PYFZX177)

摘　　要：对虚拟场景中的篡改语音进行检测,可增加语音的安全性。由于在虚拟场景中语音声纹特征无法与外界环境形成互动,使得语音声纹发生突变性变化。传统的检测方法只能通过增加可检测语音数据量对突变性进行制约,当语音信号数量过大时,出现检测效率低的问题。提出一种新的虚拟场景中篡改语音的检测方法,利用虚拟现实技术对语音检测环境建模,在分析语音变换基本原理及构造模型和变换的语音缺失帧的基础上,通过对语音信号的声道参数提取及相应的信号统计量,利用向量机选择出当语音变换发生时随之变换的特征作为分类特征并进行语音变换检测。实验结果表明,上述方法具有较高的检测准确率,实现了对语音变换的语音篡改检测,具有良好的鲁棒性。To detect the tempered voice in virtual scene can improve the voice security. The feature of voice vocal print cannot interact with external environment in virtual scene, which makes vocal print change suddenly. Traditional detection method only increases the voice data amount to restrict mutability, leading to low detection effective when the number of voice signal is too large. In this paper, we proposed a new detection method of tampered voice in virtual scene. Firstly we used the virtual reality technology to build model of voice detection environment. Then we extracted the vocal tract parameters of voice signal and corresponding signal statistics based on analyzing basic principle of voice transmission, structure model and variable absence frame. Finally we used vector machine to select the feature changing with the voice conversion as the classification feature and detected the voice conversion. The experimental results show that the method mentioned above has higher detection precision. It achieves the tampered voice detection of voice conversion and has better robustness.

关 键 词：虚拟场景 语音变换 声道参数 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]