近年来,随着社会生活水平的不断提高,人们对机器智能人声识别的要求越来越高。高斯混合—隐马尔可夫模型(Gaussian of mixture-hidden Markov model,GMM-HMM)是说话人识别研究领域中最重要的模型。由于该模型对大语音数据的建模能力不...近年来,随着社会生活水平的不断提高,人们对机器智能人声识别的要求越来越高。高斯混合—隐马尔可夫模型(Gaussian of mixture-hidden Markov model,GMM-HMM)是说话人识别研究领域中最重要的模型。由于该模型对大语音数据的建模能力不是很好,对噪声的顽健性也比较差,模型的发展遇到了瓶颈。为了解决该问题,研究者开始关注深度学习技术。引入了CNN深度学习模型研究连续语音说话人识别问题,并提出了CNN连续说话人识别(continuous speaker recognition of convolutional neural network,CSR-CNN)算法。模型提取固定长度、符合语序的语音片段,形成时间线上的有序语谱图,通过CNN提取特征序列,经过奖惩函数对特征序列组合进行连续测量。实验结果表明,CSR-CNN算法在连续—片段说话人识别领域取得了比GMM-HMM更好的识别效果。展开更多
结合指纹和声纹在识别环境中的优良互补性以及两种特征相关性不强的特点,提出一种基于指纹和声纹决策级融合识别方法。在对相应的指纹和声纹分别提取特征和识别之后,使用提出的二次比较选择模型(second comparison of choosing model,SC...结合指纹和声纹在识别环境中的优良互补性以及两种特征相关性不强的特点,提出一种基于指纹和声纹决策级融合识别方法。在对相应的指纹和声纹分别提取特征和识别之后,使用提出的二次比较选择模型(second comparison of choosing model,SCCM)在决策层上利用分数层数据信息进行辅助分类,根据类别判断是否启用声纹识别,使得双模态融合系统平均数据处理量降低了32.8%。基于FVC2002DB1指纹库和自采集声纹库进行了实验验证。实验结果显示,多模态融合可以有效解决单模态的识别限制,识别率相对单模态提升了将近5.08%和4.60%,且稳定性较高。展开更多
Equipped with multiple and unique features,a terahertz absorber exhibits great potential for use in the development of communication,military,and other fields where achieving perfect broadband absorption has always be...Equipped with multiple and unique features,a terahertz absorber exhibits great potential for use in the development of communication,military,and other fields where achieving perfect broadband absorption has always been a challenge.We present a metamaterial terahertz(THz)absorber comprising a cross-dipole patch,four symmetric square patches and an asymmetric open-loop patch with a good perfect absorption rate for TE and TM polarizations.The average absorption of more than 96%occurs in the frequency range from 2.4 THz to3.8 THz,in which the absorptance peak can reach 99.9%,as indicated by simulated results.Our design has broad potential applications in THz couplers,as well as in fields like biology and security.展开更多
文摘结合指纹和声纹在识别环境中的优良互补性以及两种特征相关性不强的特点,提出一种基于指纹和声纹决策级融合识别方法。在对相应的指纹和声纹分别提取特征和识别之后,使用提出的二次比较选择模型(second comparison of choosing model,SCCM)在决策层上利用分数层数据信息进行辅助分类,根据类别判断是否启用声纹识别,使得双模态融合系统平均数据处理量降低了32.8%。基于FVC2002DB1指纹库和自采集声纹库进行了实验验证。实验结果显示,多模态融合可以有效解决单模态的识别限制,识别率相对单模态提升了将近5.08%和4.60%,且稳定性较高。
文摘Equipped with multiple and unique features,a terahertz absorber exhibits great potential for use in the development of communication,military,and other fields where achieving perfect broadband absorption has always been a challenge.We present a metamaterial terahertz(THz)absorber comprising a cross-dipole patch,four symmetric square patches and an asymmetric open-loop patch with a good perfect absorption rate for TE and TM polarizations.The average absorption of more than 96%occurs in the frequency range from 2.4 THz to3.8 THz,in which the absorptance peak can reach 99.9%,as indicated by simulated results.Our design has broad potential applications in THz couplers,as well as in fields like biology and security.
