改进的梅尔倒谱系数在低空飞行器特征提取中的应用*

梅尔倒谱系数特征提取技术依据人耳的感知特性将声信号从线性频域转换到梅尔域,在语音识别中得到广泛应用。该文将梅尔倒谱系数技术用于小型低空飞行器的声信号特征提取中,并针对螺旋桨驱动类的小型低空飞行器具有稳定的强谐波特性,对梅尔倒谱系数特征提取中使用的梅尔滤波器进行改进,通过对此类谐波处的线性频谱与梅尔谱转换曲线的斜率进行投影替换,提高滤波器对该谐波处信号的感知敏感度。仿真结果表明,使用改进的梅尔倒谱系数特征提取方法对小型低空飞行器进行特征提取时,能够得到更低的等误识率,并且在低信噪比环境中,改进的梅尔倒谱系数特征提取方法具有更好的抗噪能力。     

水下声目标的梅尔倒谱系数智能分类方法

传统水下声目标识别分类方法具有较强的人机交互特性,无法满足未来水下无人平台智能识别分类水声目标的需求。针对这一问题,提出了一种基于梅尔倒谱系数（MFCC）的水下声目标智能识别分类方法,该方法通过提取水下声目标梅尔倒谱系数特征,采用长短时记忆网络（LSTM）构建了智能识别分类模型。使用实际水声信号对该方法进行了验证,结果表明,基于梅尔倒谱系数的水下声目标智能识别分类方法能够在不依赖人工提取特征的情况下,对目标噪声进行识别分类,具备智能化识别分类能力。     

基于动态单边自相关序列和频率规整线性预测的抗噪声语音识别

提出了一种既符合人耳听觉特性又具有良好抗噪性的语音特征分析方法。首先将单边自相关函数序列进行时间方向的平滑处理,提高单边自相关函数的抗噪性,然后用平滑后的单边自相关函数序列代替原信号进行频率规整的LPC分析,最后经倒谱变换得到该特征参数。数字语音识别实验证明:利用该特征参数的语音识别系统的识别性能优于MEL倒谱系数、LPC倒谱系数等传统的语音特征参数。     

采用L1/2稀疏约束的梅尔倒谱系数语音重建方法

提出了一种利用L_1/2稀疏约束从梅尔倒谱系数重建语音时域信号方法。从梅尔倒谱系数估计语音幅度谱是一个欠定问题,现有的方法均采用幅度谱最小均方误差估计或采用L1正则化进行幅度谱的稀疏约束。相比于L₁正则化模型,L_1/2的稀疏约束特性更强,为此,本文在从梅尔倒谱系数估计语音幅度谱时引入L_1/2正则化约束,并利用求解的稀疏幅度谱估计相位谱,最后利用估计的频谱重建时域语音信号。实验结果表明,与幅度谱最小均方误差法相比,本文算法所估计出的语音信号具有更高的语音质量;在噪声环境下进行语音重建实验,与L₁正则化幅度谱估计方法相比,本文算法重建的语音质量更好,表现出更好抗噪性。      

基于仿生特征模式匹配的海洋动物声信号识别*

针对小数据量的海洋动物声信号混合识别,将声信号同态分析过程中的线性频率转换为Mel频率,得到模拟人耳听觉特性的Mel频率倒谱系数作为声信号的特征。按照声信号所属的物种建立特征模板,使用动态时间规整算法对待识别特征进行分类识别,并对特征库和识别算法进行优化。分别提取了6种鱼类、3种虾类、12种鲸类的Mel频率倒谱系数,为每个物种建立特征模板。分3次对3种、5种、6种鱼类进行识别,分别获得了100%、96.25%、94.68%的识别率。对6种鱼类、3种虾类、12种鲸类共21个物种进行混合识别,总识别率由87.56%提升至优化后的88.96%。实验结果表明,基于Mel频率倒谱系数和动态时间规整算法的海洋动物声信号混合识别能够在小数据量时获得较高的识别率,优化后的特征库和识别算法能够提升识别率。     

滚动轴承声信号特征提取和诊断试验研究

为解决振动检测方法不能有效识别低速旋转机械滚动轴承故障问题,利用声发射检测方法,建立了滚动轴承低速声发射信号采集试验装置,对模拟人工缺陷滚动轴承声发射信号进行了采集,进而对滚动轴承声发射信号进行总体平均经验模式分解,结合能量矩及相关系数法综合判断分解后各模态分量的真伪,据此提取出特征信号并做出其局部Hilbert边际谱,最后对滚动轴承各种故障模式进行诊断。试验结果表明该诊断方法能准确识别滚动轴承声发射信号故障频率,依据特征频率及幅值大小可对低速滚动轴承故障进行有效诊断。     

无人值守变电站电力设备音频监测及故障诊断系统

针对智能变电站的无人值守需求及现有故障诊断系统的不足,提出一种电力设备音频监测及故障诊断系统。根据变电站电力设备音频信号信噪比较低的特点,采用具有强鲁棒性的梅尔频率倒谱系数作为判断音频信号异常的特征参数,在此基础上根据音频特征构成多样本观测序列,并采用隐马尔科夫模型进行故障诊断,通过对比对数似然估计概率的输出值确定故障类型。该方法具有实时性较强的优势,也避免了现有故障诊断方法要求较大样本容量的缺陷。实验结果表明,该故障诊断系统具有较高的识别率和鲁棒性。     

基于深度学习的语声抑郁识别*

世界各地抑郁症患者数量不断增多,抑郁症的诊断和治疗面临着医生短缺问题,针对这一问题,提出了CNN和结合注意力机制的BLSTM特征融合模型。从特征选择和网络构架两方面进行了研究,对比了几种经典语声特征,得出梅尔倒谱系数对抑郁分类效果最好,再将梅尔倒谱系数分别送进CNN和结合注意力机制的BLSTM网络实现抑郁分类。在DAIC-WOZ数据集上进行实验,所提出的方法对语声抑郁的分类精确度达到78.06 %,F1分数达到74.68%。 关键词：抑郁识别;语声分析;分类     

倒谱参数稀疏分解下的汉语音谎言检测

为了提高汉语语音的谎言检测准确率,提出了一种对信号倒谱参数进行稀疏分解的方法。首先,采用小波包滤波器组对语音信号进行多频带划分,求得子频带对数能量并进行离散余弦变换以提取小波包频带倒谱系数,结合梅尔频率谱系数得到倒谱参数;其次,依据K-奇异值分解方法分别利用说谎和非说谎两种状态下的语音倒谱参数集训练得到过完备混合字典,在此字典上根据正交匹配追踪算法对参数集进行稀疏编码提取稀疏特征;最终进行多种分类模型下的识别实验·实验结果表明,稀疏分解方法相比传统参数降维方法具有更好的优化性能,本文推荐的稀疏谱特征最佳识别率达到78.34%,优于其他特征参数,显著提高了谎言检测识别准确率。      

融合声纹信息的能量谱图在鸟类识别中的研究

常用的梅尔倒谱系数结合高斯混合模型(MFCC+GMM)方法的鸟鸣声识别技术难适应噪声环境,模型难以收敛,且计算复杂度高。该文提出一种融合声纹信息的能量谱图的鸟类识别方法(VPS-BR),该方法利用鸟类鸣声在能量谱图上所表现的多维差异性,定量识别鸣声声纹特征。通过对分贝能量进行颜色映射得到能量谱图,提取其视觉特征所表达的声学特征,分析归纳得到鸟类特有鸣声模式。在特征提取步骤中,选用识别速度快的局部二值模式、识别鲁棒性高的方向梯度直方图两个参数表征鸟鸣声谱图的边缘声纹;在识别步骤中,用局部二值模式和方向梯度直方图两种特征分别与支持向量机、K最近邻和随机森林3种分类器算法进行两两组合构建识别模型测试。对15种原始带噪鸟类鸣声数据集进行交叉验证,VPS-BR模型的平均识别率比MFCC+GMM组合模型高出11.3%,方向梯度直方图特征与K最近邻分类器的组合模型识别率达90.5%,表现出较好的抗噪性能和识别性能。最后针对样本数据集缺乏问题,使用生成对抗网络进行图像增强,进一步将识别率提升1.48%。     

基于卷积神经网络的气体绝缘组合开关盆式 绝缘子螺栓松动检测方法*

盆式绝缘子是GIS的关键绝缘器件,它与两侧气室法兰通过螺栓进行紧固连接,当螺栓松动时会导致盆式绝缘子应力分布不均,严重时会引起绝缘子破裂,从而影响GIS运行的安全性和可靠性。文章搭建了盆式绝缘子螺栓松动超声波检测系统,以获取不同螺栓不同工况下的超声信号,基于卷积神经网络对超声信号进行特征提取,并且与BP神经网络的训练结果进行对比分析。实验结果表明,卷积神经网络可以自动提取GIS盆式绝缘子螺栓松动特征量,对十种螺栓松动工况的识别准确率达到100%,相比于BP神经网络具有较高的识别准确率,该方法可以直接用于盆式绝缘子螺栓松动检测。     

基于声发射和GAN-CNN的铝合金管道法兰 连接松动泄漏检测*

面向管道法兰连接松动引起的泄漏检测需求,为解决数据样本不足和减少特征指标手动选取的繁琐环节。本文,考虑到生成性对抗网络（GAN）作为数据扩充工具,已被证明能够生成与真实数据相似的样本数据。同时,卷积神经网络（CNN）作为一种深度学习方法,为自动提取数据的特征提供了一种有效的方法。开展了基于GAN和CNN的铝合金管道法兰连接松动泄漏检测研究。首先,搭建管道泄漏标定和数据采集实验台,利用声发射技术获取不同等级的原始泄漏信号。其次,采用GAN生成样本数据扩充原始数据。同时,为了评估生成模型的性能,引入统计特评估生成质量。最后,将生成的样本数据与原始数据设置为不同训练集,基于卷积神经网络构建智能分类检测模型,应用于管道泄漏检测。同时,分类检测结果与小样本智能分类方法SVM进行了比较,实验结果表明,基于GAN和CNN构建的智能分类模型可显著提高管道法兰连接松动泄漏检测精度。     

Bolt axial stress measurement based on a mode-converted ultrasound method using an electromagnetic acoustic transducer

A method is proposed to measure the stress on a tightened bolt using an electromagnetic acoustic transducer (EMAT). A shear wave is generated by the EMAT, and a longitudinal wave is obtained from the reflection of the shear wave due to the mode conversion. The ray paths of the longitudinal and the shear wave are analyzed, and the relationship between the bolt axial stress and the ratio of time of flight between two mode waves is then formulated. Based on the above outcomes, an EMAT is developed to measure the bolt axial stress without loosening the bolt, which is required in the conventional EMAT test method. The experimental results from the measurement of the bolt tension show that the shear and the mode-converted longitudinal waves can be received successfully, and the ratio of the times of flight of the shear and the mode-converted longitudinal waves is linearly proportional to the bolt axial tension. The non-contact characteristic of EMAT eliminates the effect of the couplant and also makes the measurement more convenient than the measurement performed using the piezoelectric transducer. This method provides a promising way to measure the stress on tightened bolts.     

Modal decomposition method for acoustic impedance testing in square ducts

Accurate duct acoustic propagation models are required to predict and reduce aircraft engine noise. These models ultimately rely on measurements of the acoustic impedance to characterize candidate engine nacelle liners. This research effort increases the frequency range of normal-incidence acoustic impedance testing in square ducts by extending the standard two-microphone method (TMM), which is limited to plane wave propagation, to include higher-order modes. The modal decomposition method (MDM) presented includes four normal modes in the model of the sound field, thus increasing the bandwidth from 6.7 to 13.5 kHz for a 25.4 mm square waveguide. The MDM characterizes the test specimen for normal- and oblique-incident acoustic impedance and mode scattering coefficients. The MDM is first formulated and then applied to the measurement of the reflection coefficient matrix for a ceramic tubular specimen. The experimental results are consistent with results from the TMM for the same specimen to within the 95% confidence intervals for the TMM. The MDM results show a series of resonances for the ceramic tubular material exhibiting a monotonic decrease in the resonant peaks of the acoustic resistance with increasing frequency, resembling a rigidly-terminated viscous tube, and also evidence of mode scattering is visible at the higher frequencies.     

ITER�ݳ�����֧�ű����Ƶ���˨����̽��

用解析法分析了螺栓的受力情况,计算出了螺栓的预紧力,设计了各种螺栓的参数,介绍了变更设计中用超级螺栓来防松的问题,提出了用全尺寸的螺栓模型和支撑结构的装配模型进行实际工程测试研究来增加设计的可靠性,为ITER纵场磁体支撑的螺栓结构设计提供了工程参考。     

An acoustic system for the individual recognition of insects

Research into acoustic recognition systems for insects has focused on species identification rather than individual identification. In this paper, the feasibility of applying pattern recognition techniques to construct an acoustic system capable of automatic individual recognition for insects is investigated analytically and experimentally across two species of Orthoptera. Mel-frequency cepstral coefficients serve as the acoustic feature, and α-Gaussian mixture models were selected as the classification models. The performance of the proposed acoustic system is promising and displays high accuracy. The results suggest that the acoustic feature and classifier method developed here have potential for individual animal recognition and can be applied to other species of interest.     

ITER纵场磁体支撑变更设计的螺栓问题探讨

用解析法分析了螺栓的受力情况,计算出了螺栓的预紧力,设计了各种螺栓的参数,介绍了变更设计中用超级螺栓来防松的问题,提出了用全尺寸的螺栓模型和支撑结构的装配模型进行实际工程测试研究来增加设计的可靠性,为ITER纵场磁体支撑的螺栓结构设计提供了工程参考。     

Interaction of Rayleigh waves with a rib attached to a plate

An analysis of the interaction of a pulse of Rayleigh waves on a plate with a rectangular flange or rib attached in the far field of its acoustic emission (AE) source is presented. A finite difference (FD) model is used to give numerical visualizations of the development of the complete wave-field propagating through the structure and also to give displacement data for selected points. An ‘energy partition model’, for Rayleigh wave propagation across the rib has been developed and is described. Ray theory wave arrival time calculations are given for the propagation of various mode-converted components transmitted across the flange. A series of experimental measurements of AE wave propagation across a T-butt welded rib on a plate were made and these are compared with the model predictions. The practical implications of these findings for acoustic emission monitoring of large structures of complex shape are discussed.     

HDP-HSMM的磨削声发射砂轮钝化状态识别*

在高精度金属材料磨削加工中,刀具即砂轮的状态对加工效率和加工质量具有重要的影响。钝化程度较高的砂轮不适于加工精密工件,需提前预警并修整更换砂轮。该文提出一种通过磨削声发射信号来检测砂轮钝化状态的方法。首先,对于采集到的信号进行小波软阈值降噪。然后,将其分割成多个有重叠的帧,并提取每帧信号的8个特征组成声发射数据集。最后,通过分层Dirichlet过程-隐半马尔可夫模型来建立声发射数据集和不同的砂轮钝化状态之间的非线性关系,旨在识别砂轮钝化状态。结果表明,上述检测方法能有效识别砂轮的不同钝化状态并能对整个加工过程中的砂轮钝化程度进行自动划分,其在测试数据集上的准确率达到93.7%,可以为实际工业应用提供理论指导。     

A cochlear model for acoustic emissions

Variability in cochlear emission properties among different species, particularly humans and small mammals, and within individuals in the same species, is modeled by a cochlear nonlinear transmission line. The difference between humans and animals is largely explained by a lower cochlear input impedance in human ears than in cats, gerbils, or chinchillas. Inconstancy in emission properties among individual human or animal subjects is related to structural variability among ears, which can be the result of a nonuniform connection between the outer hair cells cilia and the tectorial membrane. These structural differences are modeled by a nonuniform cochlear partition resistance along the cochlear length. The model predicts that an ear which has a uniform cochlear partition resistance and an adequate cochlear input impedance will emit acoustic distortion products (ADP), but not spontaneous acoustic emission (SAE), nor click-evoked emission (CE). Only a nonuniform cochlea emits SAE and CE in addition to enhanced ADPs. The model predictions agree quantitatively with cochlear emission data from humans and animals.