基于两种GMM-UBM多维概率输出的SVM语音情感识别*

针对GMM应用于情感识别时区分能力较弱的缺点,提出了一种将GMM与SVM有效结合的算法:基于GMM-UBM多维概率输出的SVM语音情感识别方法。该方法将GMM-UBM模型对一条语音的情感特征参数的两种多维概率输出（与特征向量同维、与GMM阶数同维）作为SVM分类器的特征参数,既利用了GMM表征数据本身统计特性的能力,又保留了SVM判决能力强的特点。在柏林情感语音库与汉语情感语料库上进行的实验结果表明,该方法在语音情感识别上的平均识别率较标准GMM方法提高1.7%3.7%。     

A ‘No Panacea Theorem’ for classifier combination

We introduce the ‘No Panacea Theorem’ (NPT) for multiple classifier combination, previously proved only in the case of two classifiers and two classes. In this paper, we extend the NPT to cases of multiple classifiers and multiple classes. We prove that if the combination function is continuous and diverse, there exists a situation in which the combination algorithm will give very bad performance. The proof relies on constructing ‘pathological’ probability density distributions that have high densities in particular areas such that the combination functions give incorrect classification. Thus, there is no optimal combination algorithm that is suitable in all situations. It can be seen from this theorem that the probability density functions (pdfs) play an important role in the performance of combination algorithms, so studying the pdfs becomes the first step of finding a good combination algorithm. Although devised for classifier combination, the NPT is also relevant to all supervised classification problems.     

Surface Tension of HFC Refrigerant Mixtures

The surface tension of refrigerant mixtures, i.e., R-410A (50 mass% R-32/50 mass% R-125), R-410B (45 mass% R-32/55 mass% R-125), R-407C (23 mass% R-32/25 mass% R-125/52 mass% R-134a), R-404A (44 mass% R-125/52 mass% R-143a/4 mass% R-134a), and R-507 (50 mass% R-125/50 mass% R-143a), has been measured and correlated in the present study. Although the first three mixtures are very important as promising replacements for R-22 in air-conditioners and heat-pumps and the last two are promising replacements for R-502, surface tension data for these mixtures were not previously available. The measurements were conducted under conditions of coexistence of the sample liquid and its saturated vapor in equilibrium. The differential capillary rise method (DCRM) was used, with two glass capillaries with inner radii of 0.3034±0.0002 and 0.5717±0.0002 mm. The temperature range covered was from 273 to 323 K, and the uncertainty of measurements for surface tensions and temperatures is estimated to be at most ±0.2 mN·m^–1 and ±20 mK, respectively. A mixing rule was selected for representing the temperature dependence of the resultant data. These data were successfully represented by a mixing rule using mass fraction based on the van der Waals correlation.     

Measuring the Specific Heat Capacity of Magnetic Fluids Using a Differential Scanning Calorimeter

The specific heat capacity of magnetic fluids was measured using a heat-flux-type differential scanning calorimeter (DSC). Magnetic fluids which contain 10 to 43 wt% ultrafine magnetite (Fe₃O₄) particles dispersed with surfactants in water or kerosene were used. The DSC was operated with the optimum heating rate (10 K·min^–1) and with equal heat capacities of sample and standard materials in the temperature range, 295 to 345 K. Synthetic sapphire (-Al₂O₃) was used as the standard reference material, and the sample pan was made from aluminum. The differences between the measured specific heat capacity values of pure water and pure copper and the recommended values were within ±3 and ±1%, respectively. The specific heat capacity data of magnetic fluids showed weak temperature dependence, but strong concentration dependence. The measured values of specific heat capacity of magnetic fluids were compared with calculated values using a mixing rule.     

Equation of State and Thermodynamic Properties of Pure D2O and D2O + H2O Mixtures in and Beyond the Critical Region

A parametric crossover model is adapted to represent the thermodynamic properties of pure D₂O in the extended critical region. The crossover equation of state for D₂O incorporates scaling laws asymptotically close to the critical point and is transformed into a regular classical expansion far from the critical point. An isomorphic generalization of the law of corresponding states is applied to the prediction of thermodynamic properties and the phase behavior of D₂O + H₂O mixtures over a wide region around the locus of vapor-liquid critical points. A comparison is made with experimental data for pure D₂O and for the D₂O + H₂O mixture. The equation of state yields a good representation of thermodynamic property data in the range of temperatures 0.8T _c(x)T1.5T _c(x) and densities 0.35_c(x)1.65_c(x).     

Accelerated EM-based clustering of large data sets

Motivated by the poor performance (linear complexity) of the EM algorithm in clustering large data sets, and inspired by the successful accelerated versions of related algorithms like k-means, we derive an accelerated variant of the EM algorithm for Gaussian mixtures that: (1) offers speedups that are at least linear in the number of data points, (2) ensures convergence by strictly increasing a lower bound on the data log-likelihood in each learning step, and (3) allows ample freedom in the design of other accelerated variants. We also derive a similar accelerated algorithm for greedy mixture learning, where very satisfactory results are obtained. The core idea is to define a lower bound on the data log-likelihood based on a grouping of data points. The bound is maximized by computing in turn (i) optimal assignments of groups of data points to the mixture components, and (ii) optimal re-estimation of the model parameters based on average sufficient statistics computed over groups of data points. The proposed method naturally generalizes to mixtures of other members of the exponential family. Experimental results show the potential of the proposed method over other state-of-the-art acceleration techniques.

Phase behavior of the system He-N2 at high pressures

Recent investigation at our institute revealed that the solid-fluid-fluid three-phase line of the system helium-nitrogen shows two quadruple points in the pressure range up to 10 GPa. Since each quadruple point is connected with four three-phase lines, the phase diagram is very complicated. We have detected the phase transitions representing solid-solid-fluid equilibria. Moreover, two lines of constant composition have been determined as a function of temperature and pressure. These results are discussed together with the implications for the phase diagram of both He-N₂ and pure nitrogen.Paper presented at the Tenth Symposium on Thermophysical Properties, June 20–23, 1988, Gaithersburg, Maryland, U.S.A.     

Absolute measurements of the thermal conductivity of mixtures of alkene-glycols with water

New absolute measurements of the thermal conductivity of ethylene and propylene glycol and their mixtures with water are presented. The measurements were performed in a tantalum-type transient hot-wire instrument at atmospheric pressure, in the temperature range 295–360 K. The overall uncertainty of the reported values is estimated to be less than ±0.5%, an estimate confirmed by measurements of the thermal conductivity of water. The mixtures with water studied have compositions of 25, 50, and 75%, by weight. A recently proposed semiempirical scheme for the prediction of the thermal conductivity of pure liquids is extended to allow the prediction of the thermal conductivity of these mixtures from the pure components, as a function of both composition and temperature.     

Effect of pressure on the solid-liquid phase equilibria of binary organic systems

Solid-liquid phase equilibria of the binary mixtures of the organic liquids have been investigated at temperatures from 278 to 343 K and pressures up to 500 MPa using a high-pressure optical vessel. The systems investigated are as follows: (1) simple eutectic systems-benzene + cyclohexane system and benzene + 2-methyl-2-propanol system; (2) eutectic systems with formation of intermolecular compounds — carbon tetrachloride + p-xylene system and carbon tetrachloride + benzene system; (3) partial solid solution system--methylnaphthalene + -methylnaphthalene system; and (4) complete solid solution system-chlorobenzene + bromobenzene system. The uncertainties of the measurements of temperature, pressure and composition are within ±0.1 K, ± 0.5 MPa, and ±0.001 mole fraction, respectively. The freezing and melting temperatures at a constant composition increase monotonously with pressure. The eutectic mixture becomes richer in the component whose temperature coefficient of the freezing pressure is larger and the eutectic temperature rises monotonously with increasing pressure in the eutectic systems. The pressure-temperature-composition relation of the solid-liquid phase equilibria can be expressed satisfactorily by an equation newly proposed.Paper presented at the Tenth Symposium on Thermophysical Properties, June 20–23, 1988, Gaithersburg, Maryland, U.S.A.     

Multidimensional single-index signal regression

In general, linearity is assumed to hold in multivariate calibration (MVC), but this may not be true. We approach the MVC problem using multidimensional penalized signal regression, which can be extended with an explicit link function between linear prediction and response and in the spirit of single-index models. As the two-dimensional surface of calibration coefficients is smoothly and generally estimated with tensor product P-splines, the unknown link function is estimated using univariate Psplines. The methods presented are grounded in penalized regression, where difference penalties are placed on the rows and columns of the tensor product coefficients, as well as on the link function coefficients, each having its own tuning parameter. An application to ternary mixture data shows that a non-linearity is present. Performance comparisons are made to standard penalized signal regression, not only demonstrating the nonlinear effect, but also improvements in external prediction.