Language model combination and adaptation using weighted finite state transducers

In speech recognition systems language model (LMs) are often constructed by training and combining multiple n-gram models. They can be either used to represent different genres or tasks found in diverse text sources, or capture stochastic properties of different linguistic symbol sequences, for example, syllables and words. Unsupervised LM adaptation may also be used to further improve robustness to varying styles or tasks. When using these techniques, extensive software changes are often required. In this paper an alternative and more general approach based on weighted finite state transducers (WFSTs) is investigated for LM combination and adaptation. As it is entirely based on well-defined WFST operations, minimum change to decoding tools is needed. A wide range of LM combination configurations can be flexibly supported. An efficient on-the-fly WFST decoding algorithm is also proposed. Significant error rate gains of 7.3% relative were obtained on a state-of-the-art broadcast audio recognition task using a history dependently adapted multi-level LM modelling both syllable and word sequences. ©2010 IEEE.

Application of the fixed point method for solution in time stepping finite element analysis using the inverse vector Jiles-Atherton model

An implementation of the inverse vector Jiles-Atherton model for the solution of non-linear hysteretic finite element problems is presented. The implementation applies the fixed point method with differential reluctivity values obtained from the Jiles-Atherton model. Differential reluctivities are usually computed using numerical differentiation, which is ill-posed and amplifies small perturbations causing large sudden increases or decreases of differential reluctivity values, which may cause numerical problems. A rule based algorithm for conditioning differential reluctivity values is presented. Unwanted perturbations on the computed differential reluctivity values are eliminated or reduced with the aim to guarantee convergence. Details of the algorithm are presented together with an evaluation of the algorithm by a numerical example. The algorithm is shown to guarantee convergence, although the rate of convergence depends on the choice of algorithm parameters. © 2011 IEEE.

A finite element model of magnetization of superconducting bulks using a solid-state flux pump

Superconductors have a bright future; they are able to carry very high current densities, switch rapidly in electronic circuits, detect extremely small perturbations in magnetic fields, and sustain very high magnetic fields. Of most interest to large-scale electrical engineering applications are the ability to carry large currents and to provide large magnetic fields. There are many projects that use the first property, and these have concentrated on power generation, transmission, and utilization; however, there are relatively few, which are currently exploiting the ability to sustain high magnetic fields. The main reason for this is that high field wound magnets can and have been made from both BSCCO and YBCO, but currently, their cost is much higher than the alternative provided by low-Tc materials such as Nb3Sn and NbTi. An alternative form of the material is the bulk form, which can be magnetized to high fields. This paper explains the mechanism, which allows superconductors to be magnetized without the need for high field magnets to perform magnetization. A finite-element model is presented, which is based on the E-J current law. Results from this model show how magnetization of the superconductor builds up cycle upon cycle when a traveling magnetic wave is induced above the superconductor. © 2011 IEEE.