偏度最大化多通道逆滤波语音去混响研究*

房间混响会降低语音质量和语音可懂度。高阶统计量是衡量非高斯性的重要参量,基于语音非高斯特性可实现语音去混响。本文提出一种基于高阶统计量的多通道语音去混响方法,该方法首次用多通道语音信号线性预测残差的三阶统计量偏度（Skewness）构造代价函数,以去混响重建信号线性预测残差的偏度最大化为目标自适应地更新逆滤波器;同时结合语音信号的产生模型,提出基于偏度准则的线性预测与房间脉冲响应逆滤波联合估计方法,进一步提高去混响算法性能。实验结果表明,该方法相较于已有的基于线性预测残差四阶统计量峰度（Kurtosis）的方法具有更好的去混响效果,且对噪声具有更强的鲁棒性。     

Approaches to characterise chromatographic column performance based on global parameters accounting for peak broadening and skewness

Peak broadening and skewness are fundamental parameters in chromatography, since they affect the resolution capability of a chromatographic column. A common practice to characterise chromatographic columns is to estimate the efficiency and asymmetry factor for the peaks of one or more solutes eluted at selected experimental conditions. This has the drawback that the extra-column contributions to the peak variance and skewness make the peak shape parameters depend on the retention time. We propose and discuss here the use of several approaches that allow the estimation of global parameters (non-dependent on the retention time) to describe the column performance. The global parameters arise from different linear relationships that can be established between the peak variance, standard deviation, or half-widths with the retention time. Some of them describe exclusively the column contribution to the peak broadening, whereas others consider the extra-column effects also. The estimation of peak skewness was also possible for the approaches based on the half-widths. The proposed approaches were applied to the characterisation of different columns (Spherisorb, Zorbax SB, Zorbax Eclipse, Kromasil, Chromolith, X-Terra and Inertsil), using the chromatographic data obtained for several diuretics and basic drugs (β-blockers).     

Non-planar core reduction of graphs

We present a reduction method that reduces a graph to a smaller core graph which behaves invariant with respect to non-planarity measures like crossing number, skewness, coarseness, and thickness. The core reduction is based on the decomposition of a graph into its triconnected components and can be computed in linear time. It has applications in heuristic and exact optimization algorithms for the non-planarity measures mentioned above. Experimental results show that this strategy reduces the number of edges by 45% in average for a widely used benchmark set of graphs.     

Decomposing data sets into skewness modes

We derive the nonlinear equations satisfied by the coefficients of linear combinations that maximize their skewness when their variance is constrained to take a specific value. In order to numerically solve these nonlinear equations we develop a gradient-type flow that preserves the constraint. In combination with the Karhunen-Loève decomposition this leads to a set of orthogonal modes with maximal skewness. For illustration purposes we apply these techniques to atmospheric data; in this case the maximal-skewness modes correspond to strongly localized atmospheric flows. We have also checked that the results are statistically significant in spite of the finite length of the data. We show how these ideas can be extended, for example to maximal-flatness modes.     

Exponential risk measure with application to UK asset allocation

In the paper the exponential risk measure of Damant and Satchell is used to formulate an investor's utility function and the properties of this function are investigated. The utility function is calibrated for a typical UK investor who would hold different proportions of equity. It is found that, for plausible parameter values, a typical UK investor will hold more equity under the assumption of non-normality of return if his utility function has the above formulation and not the standard mean-variance utility function. Furthermore, our utility function is consistent with positive skewness affection and kurtosis aversion. Some aggregate estimates of risk parameters are calculated for the typical UK investor. These do not seem well determined, raising issues of the roles of aggregation and wealth in this model.     

Tuning degree distributions: Departing from scale-free networks

Scale-free networks are characterized by a degree distribution with power-law behavior. Although scale-free networks have been shown to arise in many areas, ranging from the World Wide Web to transportation or social networks, degree distributions of other observed networks often differ from the power-law type. Data based investigations require modifications of the typical scale-free network.We present an algorithm that generates networks in which the shape of the degree distribution is tunable by modifying the preferential attachment step of the Barabási-Albert construction algorithm. The shape of the distribution is represented by dispersion measures such as the variance and the skewness, both of which are highly correlated with the maximal degree of the network and, therefore, adequately represents the influence of superspreaders or hubs. By combining our algorithm with work of Holme and Kim, we show how to generate networks with a variety of degree distributions and clustering coefficients.