HMM关键词检出系统专

关键词检出(keywordspottin)g也称词检出(wordspottin)g,是近年来语音识别中颇受重视的研究领域,可广泛应用于电话的自动接听和对话监听、语音的录入和检索等方面。文中首先介绍了关键词检出的HMM方案及其研究成果,并对其进行了仿真实验,最后指出了关键词检出尚待解决的一些问题及研究方向。     

改进的基于点过程模型的连续语音关键词识别技术

本文建立了一种基于点过程模型的连续语音关键词识别系统,该模型不同于以往的经典模型,而是将连续语音信号处理成一系列稀疏的音素点集,通过对各音素点集进行建模得到关键词模型与背景模型,再采用滑动搜索的方式来检出关键词。实验结果表明该方法在保证90%以上识别率的同时大大降低了运算复杂度,并且在具有极少量训练样本的情况下依然具有较高的识别率,具有良好的鲁棒性。     

基于点过程模型连续语音关键词检测

提出了基于点过程模型（PPM）的连续语音关键词检测方法。该方法首先利用时态模式（TRAP）特征和多层感知器（MLP）计算每个音素的帧级后验概率,在此基础上,将语音可看作多个相互独立的事件（音素）,利用泊松过程对事件建立点过程模型,最后通过计算似然比达到关键词检测目的。实验结果表明,对8kHz采样语音,关键词平均召回率和准确率分别可达69．5％和82％以上。     

基于深度信念网络的光通信网络数据异常识别研究

为提高光通信网络数据传输能力，基于深度信念网络研究光通信网络数据异常识别方法。先构建光通信网络数据传输模型，采用深度信念网络进行传输信道均衡控制，利用模糊多分类支持向量机提取数据异常特征，构建数据分类学习模型，实现对数据异常重构和关联规则挖掘，采用深度信念网络对光通信网络数据进行异常张量切片重组，用张量对多关系网络进行建模，实现对光通信网络数据异常识别。仿真结果表明，所提改进方法的能量开销仅为1.2 kJ,生命周期为55.75 h,且识别时间仅为1.0 ms,优于其余两种方法，具有更大的应用价值。     

基于深度信念网络的烟叶部位近红外光谱分类方法研究

近红外检测作为一种快速无损的检测方法得到广泛关注。但光谱中存在大量噪声以及光谱数据的高维度和非线性等特点影响了分类模型的准确率。将深信网络（DBN）的理论改进并引入光谱特征学习中,解决高维特征间非线性关系的学习问题,采用逐层训练策略和随机梯度上升法分别进行网络预训练和微调获得网络权值;并结合支持向量机（SVM）建立近红外光谱多分类模型DBN-SVM。与基于主成分分析的分类模型PCA-SVM和基于线性判别分析的LDA-SVM分类模型进行应用比较。结果表明:DBN-SVM算法能有效地学习高维数据中的内在结构和非线性关系,由该算法构建的模型具有良好的特征学习能力和分类识别能力,而且在稳健性、各类别的灵敏度和特效度也更优。     

基于SLBP深度信念网络的人脸识别研究

针对传统局部二值模式(LBP)特征提取不充分和分类器拟合的问题,提出一种基于局部纹理特征的显著局部二值模式(SLBP)和深度学习的人脸识别方法。首先,利用改进的SLBP算法提取人脸图像局部纹理特征,建立SLBP直方图;然后构建基于深度信念网络的深度学习架构,将SLBP直方图输入到深度信念网络中,采用无监督逐层训练法和有监督BP算法去训练网络,实现网络的自学习和自优化,得到网络参数;最后,利用DBN分类、识别人脸图像。仿真实验证实,所提人脸识别方法在识别率和鲁棒性方面优于传统人脸识别方法。     

基于深度信念网络的短期电力负荷预测

对电力负荷预测的原理、步骤及方法做了简要分析,对深度信念网络做了细致描述,在此基础上,提出了用深度信念网络的方法预测短期电力负荷,并做了相应的实验,深度信念网络的预测值十分逼近实际值,预测误差的绝对值范围小,为0~0.08,且误差范围波动较小,预测稳定。表明基于深度信念网络的短期电力负荷预测模型预测精准,具有很高的预测精度和预测稳定性。     

基于深度信念网络的铁路信号设备故障识别

针对现有铁路信号设备故障识别算法特征提取不准确导致正确率偏低的问题,提出了深度信念网络(DBN)的故障识别模型。该模型首先利用无监督训练方法对DBN的多个堆叠受限玻尔兹曼机(RBM)进行预训练,获得网络初始参数;然后,结合铁路信号设备识别问题,构建BP神经网络,利用有标签样本进行反向传播训练,实现网络参数微调。实验结果表明,该模型避免特征提取的人工操作,能够有效实现铁路信号设备故障的准确智能识别。     

基于深度信念网络的光通信网络数据异常识别研究

为了降低光通信网络被攻击的概率,保证光通信的安全顺畅,提出基于深度信念网络的光通信网络数据异常识别方法。利用时间-频率相结合的算法建立光通信信道模型,获取信道特征。根据信道特征密度设计数据异常特征的判断准则,利用数据挖掘聚类算法提取异常数据特征。融合BP网络和受限玻尔兹曼机网络,确立深度信念网络结构,结合隐藏层与可见层单元的概率分布情况构建数据异常识别模型,经过数据采集、特征归一化和模型微调等过程完成光通信网络数据异常识别。仿真实验表明,所提方法能够获取准确的光通信网络异常数据特征,光通信网络数据异常识别高和误报率低。     

Hierarchical deep belief networks based point process model for keywords spotting in continuous speech

Point process model keyword spotting (KWS) system has attracted considerable attentions in the areas of keyword spotting by its capacity that can generalize from a relatively small numbers of training examples. But unfortunately, the accuracy level of the point process model is not comparable with the state‐of‐the‐art KWS systems because of the poor modeling capacity of the phoneme detector, which are based on Gaussian Mixture Models. In this paper, focus on improving the performance of detector in point process model, we propose an enhanced version of point process model, which is based on hierarchical deep belief networks (DBNs). Hierarchical DBNs are used as the phoneme detector in this system, and they combine the advantages of both the DBN and the hierarchical architecture for capturing complex statistical patterns in speech while overcoming the inherent flaws of conventional hidden Markov models and multilayer layer perceptron. Experiments results on TIMIT database show that the proposed method can yield 2% improvement. Furthermore, in the case when training examples are extremely limited, it can achieve better results over state‐of‐the‐art KWS systems. Copyright © 2013 John Wiley & Sons, Ltd.     

Dysarthric speaker identification with different degrees of dysarthria severity using deep belief networks

Dysarthria is a degenerative disorder of the central nervous system that affects the control of articulation and pitch; therefore, it affects the uniqueness of sound produced by the speaker. Hence, dysarthric speaker recognition is a challenging task. In this paper, a feature-extraction method based on deep belief networks is presented for the task of identifying a speaker suffering from dysarthria. The effectiveness of the proposed method is demonstrated and compared with well-known Mel-frequency cepstral coefficient features. For classification purposes, the use of a multi-layer perceptron neural network is proposed with two structures. Our evaluations using the universal access speech database produced promising results and outperformed other baseline methods. In addition, speaker identification under both text-dependent and text-independent conditions are explored. The highest accuracy achieved using the proposed system is 97.3%.     

利用深度置信网络的中文短信分类

为了提高垃圾短信的过滤效果,通过对中文短信内容和结构特点分析,提出了一种充分利用word2vec工具将短信内容转化为固定长度向量的特征提取算法。同时设计了深度置信网络进行学习和分类,实验表明其推广性能比已有报道结果提高了5%左右。     

Nonparametric belief propagation for self-localization of sensor networks

Automatic self-localization is a critical need for the effective use of ad hoc sensor networks in military or civilian applications. In general, self-localization involves the combination of absolute location information (e.g., from a global positioning system) with relative calibration information (e.g., distance measurements between sensors) over regions of the network. Furthermore, it is generally desirable to distribute the computational burden across the network and minimize the amount of intersensor communication. We demonstrate that the information used for sensor localization is fundamentally local with regard to the network topology and use this observation to reformulate the problem within a graphical model framework. We then present and demonstrate the utility of nonparametric belief propagation (NBP), a recent generalization of particle filtering, for both estimating sensor locations and representing location uncertainties. NBP has the advantage that it is easily implemented in a distributed fashion, admits a wide variety of statistical models, and can represent multimodal uncertainty. Using simulations of small to moderately sized sensor networks, we show that NBP may be made robust to outlier measurement errors by a simple model augmentation, and that judicious message construction can result in better estimates. Furthermore, we provide an analysis of NBP's communications requirements, showing that typically only a few messages per sensor are required, and that even low bit-rate approximations of these messages can be used with little or no performance impact.     

Reliability optimization models for embedded systems with multiple applications

Summary and Conclusions-This paper presents four models for optimizing the reliability of embedded systems considering both software and hardware reliability under cost constraints, and one model to optimize system cost under multiple reliability constraints. Previously, most optimization models have been developed for hardware-only or software-only systems by assuming the hardware, if any, has perfect reliability. In addition, they assume that failures for each hardware or software unit are statistically independent. In other words, none of the existing optimization models were developed for embedded systems (hardware and software) with failure dependencies. For our work, each of our models is suitable for a distinct set of conditions or situations. The first four models maximize reliability while meeting cost constraints, and the fifth model minimizes system cost under multiple reliability constraints. This is the first time that optimization of these kinds of models has been performed on this type of system. We demonstrate and validate our models for an embedded system with multiple applications sharing multiple resources. We use a Simulated Annealing optimization algorithm to demonstrate our system reliability optimization techniques for distributed systems, because of its flexibility for various problem types with various constraints. It is efficient, and provides satisfactory optimization results while meeting difficult-to-satisfy constraints.     

An algorithm for configuring embedded networks in reconfigurable telecommunication networks

This paper considers a problem of configuring logical networks by employing a self-planning facility in a telecommunication network carrying voice-grade calls to make the least-cost configuration where the involved system cost includes hop cost and lost-call traffic cost. The hop cost depends on the number of self-planning facilities included on routing path connecting the associated node pairs, while the lost-call traffic cost is incurred due to link capacities. The configuration problem is analyzed through dimensioning and routing on a reconfigurable network in a mixed 0/1 nonlinear programming approach for which lower bounds are found by Lagrangian relaxation embedded in a hybrid search procedure for the associated dual problem. Heuristic solution procedures are exploited and their efficiencies are tested with various numerical examples.     

基于深度卷积神经网络的红外过采样扫描图像点目标检测方法

针对红外过采样扫描成像特点,提出一种基于深度卷积神经网络的红外点目标检测方法.首先,设计回归型深度卷积神经网络以抑制扫描图像杂波背景,该网络不含池化层,输出的背景抑制图像尺寸与输入图像一致;其次,对抑制后的图像进行门限检测,提取候选目标小区域原始数据;最后,将候选目标区域数据依次输入分类型深度卷积神经网络以进一步判别目标、剔除虚警.生成大量过采样训练数据有效训练两个深度网络.结果表明,在不同杂波背景下,该方法在目标信杂比增益、检测概率、虚警概率和运算时间等方面,均优于典型红外小目标检测方法,适用于红外过采样扫描系统的点目标检测.     

Controllably mobile infrastructure for low energy embedded networks

We discuss the use of mobility to enhance network performance for a certain class of applications in sensor networks. A major performance bottleneck in sensor networks is energy since it is impractical to replace the batteries in embedded sensor nodes post-deployment. A significant portion of the energy expenditure is attributed to communications and, in particular, the nodes close to the sensor network gateways used for data collection typically suffer a large overhead as these nodes must relay data from the remaining network. Even with compression and in-network processing to reduce the amount of communicated data, all the processed data must still traverse these nodes to reach the gateway. We discuss a network infrastructure based on the use of controllably mobile elements to reduce the communication energy consumption at the energy constrained nodes and, thus, increase useful network lifetime. In addition, our approach yields advantages in delay-tolerant networks and sparsely deployed networks. We first show how our approach helps reduce energy consumption at battery constrained nodes. Second, we describe our system prototype, which utilizes our proposed approach to improve the energy performance. As part of the prototyping effort, we experienced several interesting design choices and trade-offs that affect system capabilities and performance. We describe many of these design challenges and discuss the algorithms developed for addressing these. In particular, we focus on network protocols and motion control strategies. Our methods are tested using a practical system and do not assume idealistic radio range models or operation in unobstructed environments.     

Adaptive cascade networks for deep nulling

Although the analogy between array antennas and tapped delay line filters is well established, there is nothing in antenna theory that is quite the same as the simple cascade connection of two or more filter sections. By applying adaptive techniques, however, it is possible to set up a cascade beam forming network having most of the advantages of the cascade filter connection. In particular, it is possible to provide deep nulls for the suppression of powerful jammers despite the presence of unavoidable errors in the parameter settings. Our calculations indicate that a subsidiary advantage is the fact that the cascade network will settle more rapidly than a conventional adaptive network when the eigenvalues are widely separated.