Advances in speech transcription at IBM under the DARPA EARS program

This paper describes the technical and system building advances made in IBM's speech recognition technology over the course of the Defense Advanced Research Projects Agency (DARPA) Effective Affordable Reusable Speech-to-Text (EARS) program. At a technical level, these advances include the development of a new form of feature-based minimum phone error training (fMPE), the use of large-scale discriminatively trained full-covariance Gaussian models, the use of septaphone acoustic context in static decoding graphs, and improvements in basic decoding algorithms. At a system building level, the advances include a system architecture based on cross-adaptation and the incorporation of 2100 h of training data in every system component. We present results on English conversational telephony test data from the 2003 and 2004 NIST evaluations. The combination of technical advances and an order of magnitude more training data in 2004 reduced the error rate on the 2003 test set by approximately 21% relative-from 20.4% to 16.1%-over the most accurate system in the 2003 evaluation and produced the most accurate results on the 2004 test sets in every speed category.     

Surface geometry of slide bearings after percussive burnishing

A review of literature about the effect of oil pockets on improvement of sliding elements tribological performance as well as about the changes of surface topography during “zero-wear” process is shown. The paper presents also the results of experimental investigations done in the Department of Manufacturing Processes and Production Organisation of Rzeszow University of Technology, connected with the creation of oil pockets on sliding surfaces. In order to simulate a deterministic surface a program for the visualisation of pits was written. The procedures for assessment of the oil pocket size of specific shape and oil pockets coverage are presented. The tendencies of changes of surface topography and oil pockets dimensions during “zero-wear” process are also described.     

Silicon‐containing heterocyclic polymers and thin films made therefrom

Thin films, in the range of tens of micrometers thickness, have been prepared by casting onto glass plates the chloroform or N‐methylpyrrolidone solutions of polyimides or poly(imide‐amide)s containing silicon and phenylquinoxaline units in the main chain. The polymers have been synthesized by solution polycondensation reaction of aromatic diamines having preformed phenylquinoxaline rings with bis(3,4‐dicarboxyphenyl)dimethylsilane dianhydride or with a diacid chloride resulting from the reaction of this dianhydride with p‐aminobenzoic acid. The polymers were easily soluble in polar aprotic solvents and showed high thermal stability. The free‐standing films exhibited good mechanical properties with tensile strengths in the range of 48–86 MPa, tensile modulus in the range of 1.25–2.22 GPa and elongation at break in the range of 3–37%. Electrical insulating properties of some polymer films were evaluated on the basis of dielectric constant and dielectric loss and their variation with frequency and temperature. The values of the dielectric constant at 10 kHz were in the range of 2.94–3.08 for polyimides and 3.89–4.49 for poly(imide‐amide)s. © 2006 Wiley Periodicals, Inc. J Appl Polym Sci 102: 3062–3068, 2006     

<Emphasis Type="Italic">In situ</Emphasis> monitoring of solid fat content by means of pulsed nuclear magnetic resonance spectrometry and ultrasonics

An ultrasonic technique was developed to study the crystallization process of edible fats on-line. A chirp wave was used instead of the conventional pulser signal, thus achieving a higher signal-to-noise ratio. This enabled measurements to be made in concentrated systems [≈20% solid fat content (SFC)] through a 8.11-cm thick sample without significant signal loss. Fat samples were crystallized at 20, 25, and 30°C at a constant agitation rate of 400 rpm for 90 min. The crystallization process was followed by ultrasonic spectroscopy and a low-resolution pulsed nuclear magnetic resonance spectrometer. Specific relationships were found between ultrasonic parameters [integrated response, time of flight (TF), and full width half maximum] and SFC. TF, which is an indirect measurement of the ultrasonic velocity (v), was highly correlated to SFC (r ²>0.9) in a linear fashion (v=2.601 SFC+1433.0).     

工艺对C/SiC复合材料结构与性能的影响

介绍一种化学气相渗透与先驱体浸渍裂解联用(CVI-PIP)的工艺制备碳毡增强SiC复合材料,采用SEM分析复合材料的显微结构,采用三点弯曲法测试复合材料的力学性能,结果显示在复合材料致密化过程早期,CVI工艺致密化效率明显高于PIP工艺;与完全采用PIP工艺制备C/SiC复合材料相比,采用CVI-PIP工艺可提高复合材料的致密化效率和致密化程度,复合材料残留孔隙率从18.86%下降到5.45%;相应的,C/SiC复合材料的抗弯强度与弹性模量分别从66.43 MPa和38.43 GPa增加到112.16 MPa和68.49 GPa;采用CVI-PIP联用工艺同时能够增加复合材料与其表面CVD涂层的结合性能.     

基于大规模MIMO网络的低复杂度块对角化算法

在多用户MIMO系统下行链路中,块对角化（Block diagonalization, BD）预编 码算法的和速率性能要优于匹配滤波算法（Matched filter, MF）和迫零算法（Zero forc i ng, ZF）。然而,传统的BD算法利用矩阵分解来构造除当前用户的其他所有用户信道的零空 间,需要O(N2)浮点运算次数（Float point operations, FLOPs）。当 基站的天线数N趋向于大规模时,BD算法计算复杂度巨大。本文 提出一种基于投影子方法构造其他用户合成信道的零空间的BD算法,该算法仅需O( N)FLOPs。仿真表明：同传统的BD算法相比,本文所提出的低复杂度BD算法显著地降低 了实现复杂度,而和速率性能损失微小,仍然优于MF和ZF,并且当N趋 向于大规模时,它的和速率性能趋向于传统的BD算法和SVD算法。     

考虑天然气混氢的园区综合能源系统电制氢优化配置

电制氢和天然气混氢技术在促进可再生能源消纳、降低系统碳排放量方面具有良好的理论研究和工程应用前景。面向含高比例可再生能源的园区综合能源系统,提出一种计及天然气混氢及跨季节存储的电制氢优化配置方法。首先梳理了含氢园区综合能源系统的运行框架和能量流动关系,建立园区内部能源生产、转换与存储设备的数学模型,其次以设备的年化投资成本、园区综合能源系统的年度运行成本和碳交易成本最优为目标,提出电制氢优化配置模型。最后通过算例分析表明电制氢及天然气混氢技术的引入可提升可再生能源的消纳能力,降低系统的整体经济成本和碳排放量,并分析了电解槽投资成本、混氢体积分数上限以及经济性和低碳性成本权重系数变化对规划运行结果的影响。关键词：园区综合能源系统;电制氢;天然气混氢;碳交易;跨季节储氢;优化配置 中图分类号：TM732     

Batch process monitoring in the original measurement's space

Quality control and safety related issues have become more and more important in industrial production of high added value products and chemical specialities during last years. In this regard, many successful applications of multivariate statistical process control (MSPC) for monitoring and diagnosis of batch processes have been presented. It is a common industrial practice to monitor the batch progress by exploiting the information contained in a historical database of successful batches using projection techniques such as principal components analysis (PCA), partial least squares (PLS) and independent component analysis (ICA). In this work, a new MSPC strategy for batch process monitoring is presented. Its distinctive feature is that it works in the space of the original variables. The technique uses only the T²-statistic for detection and identification purposes. The identification of the set of observations that signal the fault is accomplished by decomposing the T²-statistic as a unique sum of each variable contribution. Performance comparisons among the proposed strategy and the most popular PCA-based approaches are carried out by simulation of polymerization and penicillin cultivation batch processes. Results show that the new approach can be successfully applied to monitor this kind of processes since it works very well during both fault detection and identification stages.