基于性别识别的分类CHMM语音识别

对语音识别进行了探讨,提出一种通过性别识别对连续隐马尔可夫模型（CHMM）分类的方法,在此基础上进行语音识别。首先,通过计算性别判定语音信号的Mel频率倒谱系数（MFCC）使用CHMM对说话人性别进行识别,然后再根据不同性别使用分类CHMM进行语音识别。最后通过实验验证了方法的有效性。     

基于CFCC和相位信息的鲁棒性说话人辨识

传统的说话人识别中,人们往往认为人耳对相位信息不敏感而忽略了相位信息对语音识别的影响。为了验证相位信息对说话人识别的影响,提出了一种提取相位特征参数的方法。分别在纯净语音和带噪语音条件下,基于高斯混合模型,通过将相位特征参数与耳蜗倒谱系数（CFCC）相结合,研究了相位信息对说话人辨识性能的影响。实验结果标明：相位信息在说话人识别中也有着重要的作用,将其应用于说话人辨识系统,可明显提高系统的识别率和鲁棒性。     

基于说话人识别的IVR验证系统

信息时代,信息的安全性和验证的便利性是各大公司和系统需要考虑的首要点和重点。基于说话人识别的IVR验证系统,通过说话人识别组件分析用户的语音信息,并给出匹配相似度,再辅以验证几组安全问题,可以大大提高验证的准确性和缩短验证流程的时间,具有很好的实际应用价值。     

基于唇动的说话人识别技术

针对说话人识别技术多基于语音的现状,文章提出了一种新颖的基于唇动的说话人识别技术。通过离散余弦变换,从说话人讲话时的图像序列提取那些既反映说话人嘴部生理特性也反映了说话人唇动的行为特性的视觉特征。基于这些特征,为说话人建立静态-动态混合模型,其中使用半连续隐马尔可夫模型为说话人建立动态模型。在一个小型的视觉语料库上,我们分别对说话人辨认系统和确认系统进行实现。对说话人辨认系统,其文本有关与文本无关模式的正确率分别达到了100%和99.7%;对说话人确认系统,文本有关与文本无关模式的等错误率分别为0.09%与0.33%。     

基于HMM的性别识别

进行男女生识别的方法有很多种,如GMM,VQ等,该文提出了基于HMM进行说话人性别识别的方法,该方法通过计算语音信号的Mel频率倒谱系数(MFCC)并使用隐马尔可夫模型(HMM)进行性别识别。在实验室环境下,对50个不同说话人(其中男女说话人各占一半)的语音文件采用该方法与基于VQ的方法进行比较实验,从实验方法和实验结果方面得出结论:HMM的方法更加简单易行,识别率更高。对于实验的语音材料,采用HMM的方法识别率可以达到100%。     

基于深度学习的反语音欺骗模型

自动说话人验证（Automatic Speaker Verification,ASV）通过接收说话人的语音来提取说话人的语音特征，并在已经训练的模型上进行验证，最终识别输入语音的真伪，并判断说话人的身份。随着语音欺骗技术的发展，为保护ASV系统的安全性，需要提升反语音欺骗模型的性能。因此，文章提出基于深度学习的反语音欺骗模型。实验结果表明，该模型能够有效提高ASV系统的安全性。     

基于知识蒸馏和生成对抗网络的远场语音识别

为了进一步利用近场语音数据来提高远场语音识别的性能,提出一种基于知识蒸馏和生成对抗网络相结合的远场语音识别算法.该方法引入多任务学习框架,在进行声学建模的同时对远场语音特征进行增强.为了提高声学建模能力,使用近场语音的声学模型（老师模型）来指导远场语音的声学模型（学生模型）进行训练.通过最小化相对熵使得学生模型的后验概率分布逼近老师模型.为了提升特征增强的效果,加入鉴别网络来进行对抗训练,从而使得最终增强后的特征分布更逼近近场特征.AMI数据集上的实验结果表明,该算法的平均词错误率（WER）与基线相比在单通道的情况下,在没有说话人交叠和有说话人交叠时分别相对下降5.6%和4.7%.在多通道的情况下,在没有说话人交叠和有说话人交叠时分别相对下降6.2%和4.1%.TIMIT数据集上的实验结果表明,该算法获得了相对7.2%的平均词错误率下降.为了更好地展示生成对抗网络对语音增强的作用,对增强后的特征进行了可视化分析,进一步验证了该方法的有效性.     

基于情感语音的差异检测与剔除的说话人识别

提出了一种基于情感语音的差异检测与剔除的说话人识别方法,克服了前人的方法中需要在训练时提供测试说话人的情感语音或者需要在测试时提供测试语音的情感状态信息给系统的使用带来的不便性,并在识别性能上比传统的ASR系统提高4.7%。     

基于本征音因子分析的短时说话人识别

提出了一种基于本征音因子分析的文本无关的说话人识别方法.它解决了训练语音与测试语音均很短的情况下,传统的基于最大后验概率准则的混合高斯模型无法建立稳定的说话人模型问题.首先利用期望最大化算法在开发集上训练出说话人的本征音载荷矩阵,在说话人模型建模时通过将短时语音数据向本征音空间的降维映射来得到模型参数.实验结果表明,在NIST SRE 2006数据库中的10 s训练语音-10 s测试语音任务中,在传统的混合高斯模型的基线系统上,通过采用本征音因子分析的方法可以使系统等错误率降低18%.     

信道失配环境下鲁棒说话人识别

目前说话人识别系统在理想环境下识别率已可达90%以上,但在实际通信环境下识别率却迅速下降.本文对信道失配环境下的鲁棒说话人识别进行研究.首先建立了一个基于高斯混合模型(GMM)的说话人识别系统,然后通过对实际通信信道的测试和分析,提出了两种改进方法.一是由实测数据建立了一个通用信道模型,将干净语音经通用信道模型滤波后再作为训练语音训练说话人模型;二是通过对比实测信道﹑理想低通信道及语音梅尔倒谱系数(MFCC)的特点,提出合理舍去语音第一﹑二维特征参数的方法.实验结果表明,通过处理后,系统在通信环境下的识别率提升了20%左右,与传统的倒谱均值减(CMS)方法相比,识别率提高了9%-12%.     

基于遗传算法和灰色关联分析的击键特征识别算法

基于用户击键特征的身份认证比传统的基于口令的身份认证方法有更高的安全性，现有研究方法中基于神经网络、数据挖掘等算法计算复杂度高，而基于特征向量、贝叶斯统计模型等算法识别精度较低。为了在提高识别精度的同时有效降低计算复杂度，在研究现有算法的基础上提出了一种基于遗传算法与灰色关联分析的击键特征识别算法。该算法利用遗传算法根据用户训练样本确定表征用户击键特征的标准特征序列，通过对当前用户击键特征序列与标准特征序列进行灰色关联分析实现用户身份认证。实验结果表明，该算法识别精度达到神经网络、支持向量机等算法的较高水平，错误拒绝率与错误接受率分别为0%与1.5%。且计算复杂度低，与基于特征向量的算法相近。     

基于图分析和支持向量机的企业网异常用户检测

在企业网络中,若其内部的攻击者获得了用户的身份认证信息,其行为与正常用户将很难区分;而目前研究对于企业网中的异常用户检测方法比较单一,召回率不高。用户的认证活动信息直接反映了用户在网络中与各类资源或人员的交互,基于此,提出一种利用用户认证活动信息来检测网络中异常用户的方法。该方法利用用户的认证活动生成用户认证图,之后基于图分析方法提取认证图中的属性,如图的最大连通组件的大小、孤立认证的数量等,这些属性反映了用户在企业网中的认证行为特征。最后利用有监督的支持向量机（SVM）对提取到的图属性进行建模,以此来间接识别和检测网络中的异常用户。在提取了用户图向量之后,具体对训练集和测试集、惩罚参数、核函数取不同值的情况进行了分析。通过对这些参数的调节,召回率、精确率和F1-Score均达到80%以上。实验数据表明,该方法能够有效检测企业网络中的异常用户。     

基于指纹识别的云存储身份认证系统设计

针对WEB 云存储系统现有"静态口令+实时数据”的身份认证模式存在安全性能较低的问题,设计了基于指纹 识别技术的身份认证系统。该设计方案采用指纹识别的认证模式对云存储用户进行身份认证,提高了身份认证的安全性、可 靠性,解决了云存储用户账号非法访问、篡改等问题。提出了基于对称加密算法、非对称加密算法的混合加密算法,并将其作 为身份认证协议,有效提高了认证的效率,实现了海量数据的高效传输。实验结果表明,该方案在云存储身份认证系统中得到 很好的应用。     

Implicit Continuous User Authentication for Mobile Devices based on Deep Reinforcement Learning

The predominant method for smart phone accessing is confined to methods directing the authentication by means of Point-of-Entry that heavily depend on physiological biometrics like, fingerprint or face. Implicit continuous authentication initiating to be loftier to conventional authentication mechanisms by continuously confirming users’ identities on continuing basis and mark the instant at which an illegitimate hacker grasps dominance of the session. However, divergent issues remain unaddressed. This research aims to investigate the power of Deep Reinforcement Learning technique to implicit continuous authentication for mobile devices using a method called, Gaussian Weighted Cauchy Kriging-based Continuous Czekanowski’s (GWCK-CC). First, a Gaussian Weighted Non-local Mean Filter Preprocessing model is applied for reducing the noise present in the raw input face images. Cauchy Kriging Regression function is employed to reduce the dimensionality. Finally, Continuous Czekanowski’s Classification is utilized for proficient classification between the genuine user and attacker. By this way, the proposed GWCK-CC method achieves accurate authentication with minimum error rate and time. Experimental assessment of the proposed GWCK-CC method and existing methods are carried out with different factors by using UMDAA-02 Face Dataset. The results confirm that the proposed GWCK-CC method enhances authentication accuracy, by 9%, reduces the authentication time, and error rate by 44%, and 43% as compared to the existing methods.     

近红外指静脉图像采集系统设计

针对指静脉身份认证需求,以手指静脉图像采集系统作为研究对象,设计了基于单侧光源与反射镜面相结合红外光源可调控的指静脉图像采集系统。研究了LED光源位置与角度对静脉图像质量的影响,提出了基于图像质量评价的指静脉认证方法,并运用实测方法进行了验证。实验结果表明:可以得到与传统正面光源采集同等质量的静脉图像,具有更高的认证通过率,达到98.8%,并更易于用户使用。     

A Lightweight Approach (BL-DAC) to Secure Storage Sharing in Cloud-IoT Environments

The growing advent of the Internet of Things (IoT) users is driving the adoption of cloud computing technologies. The integration of IoT in the cloud enables storage and computational capabilities for IoT users. However, security has been one of the main concerns of cloud-integrated IoT. Existing work attempts to address the security concerns of cloud-integrated IoT through authentication, access control, and blockchain-based methods. However, existing frameworks are somewhat limited by scalability, privacy, and centralized structures. To mitigate the existing problems, we propose a blockchain-based distributed access control method for secure storage in the IoT cloud (BL-DAC). Initially, the BL-DAC performs decentralized authentication using the Quantum Neural Network Cryptography (QNNC) algorithm. IoT users and edge nodes are authenticated in the blockchain deployed by distributed Trusted Authorities (TAs) using multiple credentials. The user data is classified into sensitive and non-sensitive categories using the Enhanced Seagull Optimization (ESO) algorithm. Also, the authentication to access this data is performed by a decentralized access control method using smart contract policy. Sensitive user data is encrypted using the QNNC algorithm and stored in the private cloud. In contrast, non-sensitive data is stored in the public cloud, and IPFS is used to store data in a decentralized manner with high reliability. In addition, data security is improved by using a hierarchical blockchain which improves scalability by managing the multiple blockchains hierarchically and is lightweight using Proof of Authentication Consensus (PoAH). The BL-DAC is simulated and validated using the Network Simulator-3.26 simulation tool and validated. This work shows better results than the compared ones in terms of validation metrics such as throughput (26%), encryption time (19%), decryption time (16%), response time (15%), block validation time (31%), attack detection rate (16%), access control precision (13%), and scalability (28%).     

基于认证的网络权限管理技术

文中提出了一种基于认证的网络权限管理技术,实现了针对用户的策略和权限管理,并研究了本认证系统的可靠性,利用Bcrypt算法等技术解决了认证系统中存在的安全问题,提高了整个系统的可靠性。这项技术目前已经应用到实际工作中。     

一种基于同态加密的分布式生物特征认证协议

分布式生物特征认证系统因不依赖弱口令或硬件标识物而获得高的可靠性、安全性和便利性,但也因生物特征存在永久失效和隐私泄露的风险而面临更多的安全威胁.基于同态加密技术的生物特征认证方案允许特征向量在密文域匹配以保护向量安全和用户隐私,但也因此要在密文域执行昂贵的乘法运算,而且还可能因为向量封装不当而遭受安全攻击.在Brakerski等人同态加密方案的基础上提出了一种安全向量匹配方法,并在该方法的基础上设计了一个口令辅助的生物特征同态认证协议.该协议无需令牌等硬件标识物,注册时只需将带有辅助向量的特征模板密文和辅助向量外包存储,认证时服务器使用辅助向量匹配法完成模板向量和请求向量的相似性评估即可实现用户身份认证.基于Dolev-Yao攻击者模型变种和分布式生物特征认证系统所面临的主要攻击手段对协议进行了安全性分析,并通过和另外2个基于RLWE(learning with error over ring)同态的生物特征认证协议的对比分析,证明了新协议在隐私保护和向量匹配效率方面更具优势.     

User authentication via keystroke dynamics based on difference subspace and slope correlation degree

User authentication via keystroke dynamics remains a challenging problem due to the fact that keystroke dynamics pattern cannot be maintained stable over time. This paper describes a novel keystroke dynamics-based user authentication approach. The proposed approach consists of two stages, a training stage and an authentication stage. In the training stage, a set of orthogonal bases and a common feature vector are periodically generated from keystroke features of a legitimate user?s several recent successful authentications. In the authentication stage, the current keystroke feature vector is projected onto the set of orthogonal bases, and the distortion of the feature vector between its projection is obtained. User authentication is implemented by comparing the slope correlation degree of the distortion between the common feature vector with a threshold determined periodically using the recent impostor patterns. Theoretical and experimental results show that the proposed method presents high tolerance to instability of user keystroke patterns and yields better performance in terms of false acceptance rate (FAR) and false rejection rate (FRR) compared with some recent methods.     

A parameterized model to select discriminating features on keystroke dynamics authentication on smartphones

Nowadays, smartphones work not only as personal devices, but also as distributed IoT edge devices uploading information to a cloud. Their secure authentications become more crucial as information from them can spread wider. Keystroke dynamics is one of prominent candidates for authentications factors. Combined with PIN/pattern authentications, keystroke dynamics provide a user-friendly multi-factor authentication for smartphones and other IoT devices equipped with keypads and touch screens. There have been many studies and researches on keystroke dynamics authentication with various features and machine-learning classification methods. However, most of researches extract the same features for the entire user and the features used to learn and authenticate the user’s keystroke dynamics pattern. Since the same feature is used for all users, it may include features that express the users’ keystroke dynamics well and those that do not. The authentication performance may be deteriorated because only the discriminative feature capable of expressing the keystroke dynamics pattern of the user is not selected. In this paper, we propose a parameterized model that can select the most discriminating features for each user. The proposed technique can select feature types that better represent the user’s keystroke dynamics pattern using only the normal user’s collected samples. In addition, performance evaluation in previous studies focuses on average EER(equal error rate) for all users. EER is the value at the midpoint between the FAR(false acceptance rate) and FRR(false rejection rate), FAR is the measure of security, and FRR is the measure of usability. The lower the FAR, the higher the authentication strength of keystroke dynamics. Therefore, the performance evaluation is based on the FAR. Experimental results show that the FRR of the proposed scheme is improved by at least 10.791% from the maximum of 31.221% compared with the other schemes.