摘要
针对传统认证算法不适用于语音通讯的现状,提出一种基于人耳感知特性的认证算法,较好地满足了语音认证对鲁棒性和效率的较高要求.算法对语音提取浊音部分,基于MFCC原理提出一种新的感知特征提取算法计算AMFCC参数,然后用RM-EPN编码消除误差,最后使用Rainbow算法对序列进行签名.实验证明,该认证算法在保持篡改识别率的基础上,能够抵御20dB的白噪声干扰,对窄带噪声、信道滤波等其他正常干扰具有95%以上的通过率.同时运行效率较高.Leon3SOC平台的实验结果表明,算法能够用较少的资源实现语音的实时处理,适用于资源受限的语音通信终端.
To solve the problem that conventional authentication fail to deal with audio data,a new authentication algorithm based on perceptual characteristics is presented to meet requirement of robustness,uniqueness and efficiency.First,a voicing decision is made.Then an AMFCC algorithm are designed and applied to extracted perceptual parameters,and RM-EPN coding is used to eliminate the perturbation.Finally Rainbow algorithm is applied to sign the extracted hash.Experiment results show strong ability to resist attack of white gauss noise and other disturbances,such as narrowband noise,filtering and echo,etc.Meanwhile,the algorithm could do real-time audio processing on the platform of Leon3 SOC,which show good efficiency.
出处
《小型微型计算机系统》
CSCD
北大核心
2010年第7期1461-1465,共5页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(60772032)资助
关键词
鲁棒性
高效
感知特性
语音认证
robustness
effective
perceptual characteristic
speech authentication