Comparing ANN and GMM in a voice conversion framework

In this paper, we present a comparative analysis of artificial neural networks (ANNs) and Gaussian mixture models (GMMs) for design of voice conversion system using line spectral frequencies (LSFs) as feature vectors. Both the ANN and GMM based models are explored to capture nonlinear mapping functions for modifying the vocal tract characteristics of a source speaker according to a desired target speaker. The LSFs are used to represent the vocal tract transfer function of a particular speaker. Mapping of the intonation patterns (pitch contour) is carried out using a codebook based model at segmental level. The energy profile of the signal is modified using a fixed scaling factor defined between the source and target speakers at the segmental level. Two different methods for residual modification such as residual copying and residual selection methods are used to generate the target residual signal. The performance of ANN and GMM based voice conversion (VC) system are conducted using subjective and objective measures. The results indicate that the proposed ANN-based model using LSFs feature set may be used as an alternative to state-of-the-art GMM-based models used to design a voice conversion system.     

Finite-size effects in surface tension: Thermodynamics and the Gaussian interface model

It has been suggested by Kayser that finite-size effects associated with capillary waves might play a significant role in some surface tension measurements; for capillary rise between plates a distance D apart, an effect varying as 1/D and apparently observable in measurements, was proposed. In reconsidering this problem, one must analyze the thermodynamics of finite-size corrections to surface tension. In particular, one sees that capillary rise between plates does not measure the interfacial free energy density but, rather, a derivative of the interfacial free energy with respect to a system dimension. The quantity needed to draw definite conclusions, the finite-size residual free energy, can be calculated within the harmonic or Gaussian capillary wave model in d spatial dimensions with the aid of Poisson summation techniques and should yield the correct leading asymptotic behavior. For d=3 and experimentally relevant parameter values, the results are independent of the short-wavelength cutoff needed in the model and can be checked against the theory of conformai covariance at two-dimensional critical points. It is found that the finite-size effects in capillary-rise measurements of surface tension vary as 1/D ² (with a universal coefficient) but are too small to be seen in current experiments.Invited paper presented at the Tenth Symposium on Thermophysical Properties, June 20–23, 1988, Gaithersburg, Maryland, U.S.A.     

高斯颜色模型在瓷片图像分类中的应用

由于RGB颜色空间不能很好贴近人的视觉感知,同时也缺少对空间结构的描述,因此采用兼顾颜色信息和空间信息的高斯颜色模型以获取更全面的特征,提出了一种基于高斯颜色模型和多尺度滤波器组的彩色纹理图像分类法,用于瓷器碎片图像的分类。首先将原始图像的RGB颜色空间转换到高斯颜色模型;再用正规化多尺度LM滤波器组对高斯颜色模型的3个通道构造滤波图像,并借助主成分分析寻找主特征图,接着选取各通道的最大高斯拉普拉斯和最大高斯响应图像,与特征图联合构成特征图像组用以进行参数提取;最后以支持向量机作为分类器进行学习和分类。实验结果表明,与基于灰度的、基于RGB模型的和基于RGB_bior 4.4小波的方法相比,本文方法具有更好的分类结果,其中在Outex纹理图像库上获得的分类准确率为96.7%,在瓷片图像集上获得的分类准确率为94.2%。此方法可推广应用到其他彩色纹理分类任务。     

SVD域的图像高斯噪声强度估计

精确估计图像或视频中的噪声强度对于后续的信号处理是至关重要的先决条件。通过对含噪图像的奇异值特性的研究,提出一种精确的SVD域的图像噪声强度估计算法。该算法对噪声强度估计提出了创新的解决方法:1)利用奇异值的尾部数据进行噪声强度估计,这样达到尽可能地降低图像信息对噪声估计的干扰;2)对含噪图像加入已知强度的高斯白噪声,以计算噪声估计时需要设置的与图像内容相关的参数,因此该算法可以自适应图像的结构,能够广泛地适应各种类型的图片。实验结果表明SVD域噪声强度估计算法适用于各种图片类型,而且在极大的噪声强度范围内都能够稳定精确地估计噪声强度。     

A Random Nonstationary Pulse Train Model

A simple and mathematically tractable model of a nonstationary process is developed.The process is the sum of waves where the parameters of the waves are random. Explicit expressions for the mean and autocorrelation function at each position as a function of time are obtained.In the case of infinite time, the model evolves into a stationary process. The time-frequency distribution at each position is also obtained. An explicit example is given where the initial waves are Gaussian. The case where there is dispersion in the propagation is also discussed.     

Multi-Scale Dilated Convolutional Neural Network for Hyperspectral Image Classification

基于多尺度空洞卷积神经网络的高光谱图像分类

郑姗姗¹,刘文¹,单锐¹,赵静一²,江国乾³,张智⁴

（1. 燕山大学理学院,河北 秦皇岛 066004;2. 燕山大学机械工程学院,河北 秦皇岛 066004;3. 燕山大学电气工程学院,河北 秦皇岛 066004;4. 北京航天研究所,北京 100094）

创新点说明：

1）将图像分割方法——空洞卷积用于卷积神经网络进行高光谱图像分类,提取更加广泛、抽象的图像特征。

2）构建基于多尺度空洞卷积神经网络的高光谱图像分类方法。搭建多尺度聚合结构,在每个通道中使用快捷连接和空洞卷积结构,有效提取图像特征,避免信息丢失。

研究目的：

针对图像信息丢失问题,得到高精度的高光谱图像分类方法。

研究方法：

在Indian Pines和Pavia University数据集上,与4个已有的高光谱图像分类方法进行对比实验,比较OA, AA和Kappa值。

研究结果：

多尺度空洞卷积神经网络在Indian Pines和Pavia University数据集上OA值分别达到了99.58%,99.92%。AA值分别达到了99.57%,99.90%。Kappa分别达到了99.52%,99.89%。

结论：

1）在卷积神经网络中引入空洞卷积,可以有效避免图像信息丢失。

2）多尺度空洞卷积神经网络能提取更佳的判别性特征,实现高分类性能。

关键词：多尺度聚合;空洞卷积;高光谱图像分类;快捷连接

     

基于多任务卷积神经网络的虹膜图像质量评估方法

为了自动、准确、高效地评估采集图像的质量,设计了一个名为MTIQA的卷积神经网络。该网络能够输出与主观评价指标保持较高一致性的客观评估结果。MTIQA采用多任务学习策略,包含网络训练质量评估和失真类型分类两个任务,将两个任务的损失融合并构成新的损失函数。为了评估算法所得到的客观指标的可靠性,建立了名为SIR2019的单眼虹膜质量评估数据集,并召集志愿者进行主观实验以得到主观评价指标。在SIR2019和CASIA-Iris-Distance-Lamp数据集上的实验结果表明,该网络在虹膜图像质量评估上具有较好的可行性、准确性和鲁棒性。     

一种基于CGAN-CNN的同步电机转子绕组匝间短路故障诊断方法

由于同步电机故障样本数量较少,为解决同步电机故障诊断中普遍存在的样本不平衡问题,提出了一种基于条件生成式对抗网络(CGAN)和卷积神经网络(CNN)的同步电机转子绕组匝间短路故障诊断方法.首先,对传感器收集到的数据进行预处理,对正常样本和故障样本分别添加标签后输入CGAN中生成大量新样本,将生成的新样本与原始样本混合并划分训练集和测试集;然后,利用CNN训练平衡后的数据集,充分、精准地提取有效故障特征;最后,在输出端利用Softmax分类器输出故障分类结果.通过实验证明,与非平衡数据集相比,利用平衡数据集后的故障识别准确率十分稳定且达到99.5％以上,同时与平衡的原始样本数据相比,生成样本避免了噪声和其他干扰,故障诊断的准确率也更高.     

TESTING CHANGE-POINTS IN THE EXPLOSIVE GAUSSIAN AUTOREGRESSIVE PROCESSES

Abstract. A functional limit theorem with a particular function class and topology is derived for non-ergodic type time series. This limit theorem allows us to study the asymptotic law of the associated likelihood ratio test (LRT) statistic for testing the presence of a change in the covariance parameter in the explosive Gaussian autoregressive model. We show that the level of the LRT cannot be approximated without introducing appropriate normalization. The limit law of a particular weighted likelihood ratio test is examined through a simulation study and is compared with the well-known Kolmogorov distribution obtained in the stationary case; we conclude that for practical applications when the root is really close to unity one can use the same thresholds as in the stationary case. This procedure is applied to the study of three real time series known to be non-stationary.     

基于新型混合模型的欠定盲分离方法

针对欠定盲分离问题,提出了一种新的源恢复方法. 在时频域局部区域采用复高斯分布对源信号进行建模,将语音信号的稀疏性和局部平稳性结合在一起,提出了一种新的混合模型来描述观测信号在局部区域的概率分布.通过该模型,将每个时频点的源信号状态的判断问题转换成模型的参数估计和后验概率的计算问题,最后通过子混合矩阵的逆恢复出源信号. 实验结果表明,该方法具有很快的收敛速度,并且比已有方法具有更好的分离性能.