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.     

Philadelphia obtains useful information from its customers about taste and odour quality.

Customers are sensitive to the flavour of water. Customers evaluate drinking water based on their expectations, on experiences with their usual drinking water and on experiences with alternative waters. The Philadelphia Water Department provides one example of success in developing a better understanding of customer perceptions and attitudes about tap water taste and odour. Philadelphia found that customers do communicate in ways that water utilities can understand. Water utilities can enhance that communication and collect useful data. In addition, water utilities can characterise their tap water flavour, track it for changes and correlate changes to customer complaints.     

The prediction of solute transport in surcharged manholes using CFD.

Solute transport processes occur within a wide range of water engineering structures, and urban drainage engineers increasingly rely on modelling tools to represent the transport of dissolved materials. The models take as input representative travel time and dispersion characteristics for key system components, and these generally have to be identified via field or laboratory measurements. Computational Fluid Dynamics (CFD) has the potential to reveal the underlying hydraulic processes that control solute transport, and to provide a generic means of identifying relevant parameter values. This paper reports on a study that has been undertaken to evaluate the feasibility of utilising a CFD-based approach to modelling solute transport. Discrete phase modelling has been adopted, as this is computationally efficient and robust when compared with the time-dependent solution of the advection-dispersion equation. Simulation results are compared with published laboratory data characterising the dispersion effects of surcharged manholes, focusing specifically on an 800 mm diameter laboratory manhole for a flowrate of 0.002 m(3)/s and a range of surcharge depths. Preliminary indications are that the CFD results adequately replicate the measured downstream temporal concentration profiles, and that a threshold surcharge depth, corresponding to a change in hydraulic regime within the manhole, can also be identified.     

Measurement of the relative reduction of strip on the 1200-model temper-rolling mills in flat-rolled products shop No. 3

The temper-rolling mills in flat-rolled products shop No. 3 at the MMK has been equipped with the Vytyazhka automated system to measure the elongation of strip (ASEMS) rolled on the mill. The elongation (relative elongation) of the strip is determined through a frequency-based method of measuring the speed of the strip before and after reduction. The system transmits and displays information on the temper-rolling of each coil. The information is shown as a diagram on the monitor of an industrial computer. The reliability of the system allows it to be in continuous use. Introduction of the automated elongation measurement system makes it possible to monitor the reductions being made during temperrolling, which in turn enables the shop to reduce the number of products having properties that do not meet customers’ requirements. __________ Translated from Metallurg, No. 1, pp. 62–63, January, 2007.     

Scrap metal: The problem of its shortage and efficient use

The most important element of manufacturing processes that employ new metallurgical technologies is improving the quality of the scrap metal used in the given process. Introducing modern scrap-processing equipment is especially important in this regard. At the same time, it is also necessary to use the latest technological advances in order to reduce the oxidation loss of iron during refining in arc steelmaking furnaces.     

New technological solutions to protect the lining of blast-furnace hearths

The use of schungite as a partial replacement for coke is practiced in Russia at a number of metallurgical plants, both in the production of foundry iron (its primary use) and in making conversion pig iron. The largest amount of coke is replaced when foundry iron is being made, the replacement coefficient in this case having a value within the range 0.8–1.3. Studies that included continuous monitoring of the temperature of the carbon blocks in blast furnaces with volumes of 1719, 3200, and 5580 m³ unambiguously demonstrate that the thickness of the slag crust in blast furnaces increases with the use of schungite. Protecting the lining of the hearth through the use of schungite is also accompanied by a decrease in coke consumption. __________ Translated from Metallurg, No. 8, pp. 53–57, August, 2007.