话话人识别使用遗传RBF网络

针对ＲＢＦ网络普遍采用的一种训练算法所存在的局部最佳问题。本文将遗传算法用于ＲＢＦ网络训练过程,增强了该网络的全局寻优能力,提高了ＲＢＦ网络的模式识别性能,说话人识别实验显示,改进训练算法后,ＲＢＦ网络的说话人识别率有稍许提高。     

深浅层特征及模型融合的说话人识别

为了进一步提高说话人识别系统的性能,提出基于深、浅层特征融合及基于I-Vector的模型融合的说话人识别。基于深、浅层特征融合的方法充分考虑不同层级特征之间的互补性,通过深、浅层特征的融合,更加全面地描述说话人信息;基于I-Vector模型融合的方法融合不同说话人识别系统提取的I-Vector特征后进行距离计算,在系统的整体结构上综合了不同说话人识别系统的优势。通过利用CASIA南北方言语料库进行测试,以等错误率为衡量指标,相比基线系统,基于深、浅层特征融合的说话人识别其等错误率相对下降了54.8%,基于I-Vector的模型融合的方法其等错误率相对下降了69.5%。实验结果表明,深、浅层特征及模型融合的方法是有效的。     

基于GMM的说话人识别

通过运用mat1ab软件编程对语音信号进行了基音提取,并以MFCC、基于基音周期和MFCC的特征组合参数作为特征参数,建立了基于GMM模型的说话人识别系统。并通过识别实验得出的实验结果,发现使用基于基音周期和MFCC的特征组合参数作为特征参数,在人数为50-180人范围内,能够有效提高基于GMM说话人识别系统识别率。     

基于密集连接时延神经网络的说话人识别算法*

说话人识别技术是一项重要的生物特征识别技术。近年来,使用深度神经网络提取发声特征的说话人识别算法取得了突出成果。时延神经网络作为其中的典型代表之一已被证明具有出色的特征提取能力。为进一步提升识别准确率并节约计算资源,通过对现有的说话人识别算法进行研究,提出一种带有注意力机制的密集连接时延神经网络用于说话人识别。密集连接的网络结构在增强不同网络层之间的信息复用的同时能有效控制模型体积。通道注意力机制和帧注意力机制帮助网络聚焦于更关键的细节特征,使得通过统计池化提取出的说话人特征更具有代表性。实验结果表明,在VoxCeleb1测试数据集上取得了1.40%的等错误率（EER）和0.15的最小检测代价标准（DCF）,证明了在说话人识别任务上的有效性。     

基于改进噪声估计的谱减法应用于说话人识别

针对语音信号中存在加性噪声使MFCC的鲁棒性和识别系统的性能下降的问题,基本谱减法的引入在增强MFCC抗噪性上取得的效果有限,为了使MFCC具有更好的抗噪性,提出了一种改进算法,在谱减法的基础上引入谱熵的思想,利用谱熵值的分布逐帧进行噪声估计,可更精确地谱减去噪;实验结果表明,当语音中含有加性噪声时,与基本谱减法相比,改进谱减法的说话人识别系统抗噪性与鲁棒性更好。     

噪声环境说话人辨认初探

对于说话人辨认系统,实际使用环境与训练环境通常并不匹配,实际使用环境中常常存在各种来源的噪声,实验室训练环境中性能良好的系统在实际使用环境中其性能可能急剧下降。增强系统的抗噪能力是一个不容忽视的问题。本文试图将语音增强技术应用于说话人辨认系统,提高系统识别率和稳健性。初步实验表明,语音增强技术可显著提高系统的抗噪声能力。     

全局谱参数下的耳语说话人状态因子分析

提出了全局谱参数下的耳语说话人状态因子分析方法。首先,根据耳语听辨实验结果,提出导入唤醒度-愉悦度因子对说话人状态进行三级度量;其次,提取耳语音正弦模型、人耳听觉模型下的谱参数,结合其他短时频谱参量,进行轨迹跟踪并计算各参数的全局统计变量,作为特征参数来实现耳语说话人状态的分类。实验结果显示,正弦模型及人耳听觉模型的全局谱参数可将耳语说话人状态因子分类系统的准确率提高至90%。该分类方法及状态因子描述方案提供了耳语音说话人状态分析的有效途径。     

匹配追踪说话人自适应方法

针对现有子空间自适应方法无法确定最佳说话人子空间的问题,提出一种基于匹配追踪的说话人自适应方法。将说话人自适应视为一种高维信号的稀疏分解问题,利用本征音和参考说话人超矢量的各自优势联合构造说话人字典;依据匹配追踪原理,通过迭代优化,以后验方式确定最佳说话人子空间维数及其基矢量。引入冗余基矢量检测与去除机制以保证算法的稳定性,并通过快速递推算法得到新说话人坐标。基于汉语连续语音识别的有监督说话人自适应实验结果表明,与本征音及参考说话人加权方法相比,平均有调音节正识率相对提高了1.9%。     

深度非线性度量学习在说话人确认中的应用

将非线性度量学习(Nonlinear Metric Learning,NML)应用于说话人确认,提出了一种基于深度独立子空间分析(Independent Subspace Analysis,ISA)网络的说话人确认方法。区别于传统的线性度量学习方法,该方法使用深度独立子空间分析网络来学习一种从说话人原始空间到优化子空间的非线性显式映射,并在此基础上计算两条语音之间的相似性,以获得更好的说话人确认性能.所提方法在NIST SRE 2008数据集上进行了评估。评估结果表明,所提算法的等错误率指标相比传统的基于余弦距离打分的i-vector算法、线性判别分析(Linear Discriminant Analysis,LDA)算法、概率线性判别分析(Probabilistic Linear Discriminant Analysis,PLDA)算法分别下降了11.02%,6.40%和4.579%。     

韵律特征在概率线性判别分析说话人确认中的应用

研究韵律特征在说话人确认中的应用。将整个韵律轨迹以固定段长和段移进行片段划分,并对其进行勒让德多项式拟合从而获取连续性的韵律特征,将特征映射到总变化因子空间,并用概率线性判别分析来补偿说话人和场景的差异。在美国国家标准技术研究院2010年说话人识别评测扩展核心测试集5的基础上加入噪声构造测试集,并分别对韵律特征和传统Mel频率倒谱系数进行测试。结果显示,随着信噪比的逐渐减小,Mel频率倒谱系数性能出现大幅度下降,而韵律特征性能相对比较稳定,两种特征融合后能使系统性能得到进一步提升,等错率和最小检测错误代价相对于Mel频率倒谱系数单系统最多能分别下降9%和11%。实验表明,韵律特征应用于说话人识别中具有较强的噪声鲁棒性,且与传统的Mel频率倒谱系数存在较强的互补性。     

一种新型广义RBF神经网络在混沌时间序列预测中的研究

提出了一种新颖的广义径向基函数神经网络模型,其径向基函数(RBF)的形式由生成函数确定.然后,给出了易实现的梯度学习算法,同时为了进一步提高网络的收敛速度和网络性能,又给出了基于卡尔曼滤波的动态学习算法.为了验证网络的学习性能，采用基于卡尔曼滤波算法的新型广义RBF网络预测模型对Mackey-Glass混沌时间序列和Henon映射进行了仿真.结果表明,所提出的新型广义RBF神经网络模型能快速、精确地预测混沌时间序列,是研究复杂非线性动力系统辨识和控制的一种有效方法. 关键词： 广义径向基函数神经网络 卡尔曼滤波 梯度下降学习算法 混沌时间序列 预测     

Identification of noise sources in factory’s sound field by using genetic algorithm

Noise control is important and essential in factory, where the noise level is restricted by the Occupational Safety and Health Act. Before noise abatement being performed, the identification work in searching for the location and sound power level (SWL) of noisy sound sources is absolutely prerequisite. Several researches on new techniques of single noise control have been well addressed and developed; however, the research work on sound identification for the existing multi-noise plant is hardly found sufficient. Under the circumstance of unrecognized noises, the noise control work will expectedly be extravagant and fruitless. Therefore, the numerical approach in distinguishing noises from a multi-noise plant becomes crucial and obligatory.In this paper, the novel technique of genetic algorithm (GA) in conjunction with the method of minimized variation square will be adopted and used in the following numerical optimization. In addition, various sound monitoring systems in detecting the noise condition within the plant area will also be introduced. Before noises identification, the accuracy of mathematical model has then been proved to be in good agreements comparing to the simulated data of SoundPlan, a commercialized simulation package in sound field. Moreover, three kinds of multi-noise plants have been fully discussed and acknowledged by GA optimization. The results reveal that the relevant locations and sound power levels (SWLs) of noises can be precisely recognized. This paper surely provides a rapid methodology in the noise identification work for a multi-noise plant.     

Efficient algorithm based on neighborhood overlap for community identification in complex networks

Community structure is an important feature in many real-world networks. Many methods and algorithms for identifying communities have been proposed and have attracted great attention in recent years. In this paper, we present a new approach for discovering the community structure in networks. The novelty is that the algorithm uses the strength of the ties for sorting out nodes into communities. More specifically, we use the principle of weak ties hypothesis to determine to what community the node belongs. The advantages of this method are its simplicity, accuracy, and low computational cost. We demonstrate the effectiveness and efficiency of our algorithm both on real-world networks and on benchmark graphs. We also show that the distribution of link strength can give a general view of the basic structure information of graphs.     

Broad omnidirectional high-precision filters design using genetic algorithm

An application of the genetic algorithm in designing omnidirectional optical filters is reported in this paper. Concerning different periodic numbers and thickness ratios in the heterostructure, we gave some optimization examples and finally achieved a photonic heterostructure with very broad omnidirectional filter bandwidth as well as a very narrow transmission window. And it is found that when the normal incident beam is tilted at a negligibly small angle, the perfect transmittance peak will vanish. Hence, this heterostructure can be regarded as omnidirectional high-precision filters with potential application in optical filters, optical switches, and many other optical telecommunication areas.     

Ultrasonic flaw detection using radial basis function networks (RBFNs)

Ultrasonic flaw detection has been studied many times in the literature. Schemes based on thresholding after a previous matched filter use to be the best solution, but results obtained with this method are only satisfactory when scattering and attenuation are not considered. In this paper, we propose an alternative solution to thresholding detection method. We deal with the usage of different flaw detection methods comparing them with the proposed one. The experiment tries to determinate whether a given ultrasonic signal contains a flaw echo or not. Starting with a set of 24,000 patterns with 750 samples each one, two subsets are defined for the experiments. The first one, the training set, is used to obtain the detection parameters of the different methods, and the second one is used to test the performance of them. The proposed method is based on radial basis functions networks, one of the most powerful neural network techniques. This signal processing technique tries to find the optimal decision criterion. Comparing this method with thresholding based ones, an improvement over 25-30% is obtained, depending on the probability of false alarm. So our new method is a good alternative to flaw detection problem.     

基于径向基函数网络的激光诱导荧光特征光谱分离算法

应用激光诱导荧光技术测量水中溶解有机物(DOM)含量，具有灵敏度高、检测速度快、可遥测等优点，其中特征荧光光谱的分离在系统中占有十分重要的地位.在分析激光诱导荧光光谱特征的基础上，提出了采用径向基函数网络(RBFN)分析荧光光谱数据的数学模型，应用这种模型从荧光光谱中恢复出了激光、拉曼和DOM的荧光等光谱分量成分，从而得到了水中DOM的浓度. 关键词： 径向基函数网络 激光诱导荧光 溶解有机物     

遗传优化神经网络的水声信道盲均衡

不需要训练序列的盲均衡技术可以有效地节省水声通信带宽，消除码间干扰，提高水声通信效率和质量。以前馈神经网络（FNN）作为盲均衡器，既适用于最小相位信道，也适用于非最小相位信道，包括非线性信道，但是前馈神经网络在实际的应用中其网络拓扑结构的选取和初始权重的确定缺乏理论依据，且其训练主要依靠BP算法，存在收敛速度慢、容易陷入局部极值及“过学习”的问题。为此，本文提出了一种遗传优化神经网络的水声信道盲均衡算法（GA—BP），对前馈神经网络拓扑结构和网络权重同时优化，有效地克服了传统前馈神经网络盲均衡的缺陷，提高了前馈神经网络盲均衡的泛化性能并加强了跟踪时变信道的能力和对信道突变的适应能力。水池试验结果证明了文中提出的遗传优化神经网络水声信道盲均衡算法的有效性，与直接前馈神经网络盲均衡相比较，均衡性能明显得到了提高。     

Toward a physical basis of attention and self-regulation

The concept of self-regulation is central to the understanding of human development. Self-regulation allows effective socialization and predicts both psychological pathologies and levels of achievement in schools. What has been missing are neural mechanisms to provide understanding of the cellular and molecular basis for self-regulation. We show that self-regulation can be measured during childhood by parental reports and by self-reports of adolescents and adults. These reports are summarized by a higher order factor called effortful control, which reflects perceptions about the ability of a given person to regulate their behavior in accord with cultural norms. Throughout childhood effortful control is related to children's performance in computerized conflict related tasks. Conflict tasks have been shown in neuroimaging studies to activate specific brain networks of executive attention. Several brain areas work together at rest and during cognitive tasks to regulate competing brain activity and thus control resulting behavior. The cellular structure of the anterior cingulate and insula contain cells, unique to humans and higher primates that provide strong links to remote brain areas. During conflict tasks, anterior cingulate activity is correlated with activity in remote sensory and emotional systems, depending upon the information selected for the task. During adolescence the structure and activity of the anterior cingulate has been found to be correlated with self-reports of effortful control.Studies have provided a perspective on how genes and environment act to shape the executive attention network, providing a physical basis for self-regulation. The anterior cingulate is regulated by dopamine. Genes that influence dopamine levels in the CNS have been shown to influence the efficiency of self-regulation. For example, alleles of the COMT gene that influence the efficiency of dopamine transmission are related to the ability to resolve conflict. Humans with disorders involving deletion of this gene exhibit large deficits in self-regulation. Alleles of other genes influencing dopamine and serotonin transmission have also been found to influence ability to resolve conflict in cognitive tasks. However, as is the case for many genes, the effectiveness of COMT alleles in shaping self-regulation depends upon cultural influences such as parenting. Studies find that aspects of parenting quality and parent training can influence child behavior and the efficiency of self-regulation.During development, the network that relates to self-regulation undergoes important changes in connectivity. Infants can use parts of the self-regulatory network to detect errors in sensory information, but the network does not yet have sufficient connectivity to organize brain activity in a coherent way. During middle childhood, along with increased projection cells involved in remote connections of dorsal anterior cingulate and prefrontal and parietal cortex, executive network connectivity increases and shifts from predominantly short to longer range connections. During this period specific exercises can influence network development and improve self-regulation. Understanding the physical basis of self-regulation has already cast light on individual differences in normal and pathological states and gives promise of allowing the design of methods to improve aspects of human development.     

ZnO thin film characterization by X-ray reflectivity optimization using genetic algorithm and Fourier transformation

Zinc oxide (ZnO) thin film was fabricated by sol-gel spin coating method on glass substrate. X-ray reflectivity (XRR) and its optimization have been used for characterization and extracting physical parameters of the film. Genetic algorithm (GA) has been applied for this optimization process. The model independent information was needed to establish data analyzing process for X-ray reflectivity before optimization process. Independent information was exploited from Fourier transform of Fresnel reflectivity normalized X-ray reflectivity. This Fourier transformation (Auto Correlation Function) yields thickness of each coated layer on substrate. This information is a keynote for constructing optimization process. Specular X-ray reflectivity optimization yields structural parameters such as thickness, roughness of surface and interface and electron density profile of the film. Acceptable agreement exists between results obtained from Fourier transformation and X-ray reflectivity fitting.     

Multi-objective optimisation of lidar parameters for forest-fire detection on the basis of a genetic algorithm

A numerical model for the estimation of the laser pulse energy required for forest-fire detection with a predefined signal-to-noise ratio by an eyesafe lidar operating at wavelength of 1.54 μm is developed. This model is coupled with a genetic algorithm, which is used for multi-objective optimisation of the fitness function (weighted sum of laser price, telescope price, and surveillance time) using telescope diameter and laser beam divergence as independent variables.