Bernhard Rüber

Context-dependent probability adaptation in speech understanding

Abstract By using context-dependent concept language models, we can significantly improve the performance of the speech understanding component of a dialogue system. We use several different modelling methods, such as bigrams, concept set probability estimations using graph theory and combinations of these. The contexts are selected both manually and automatically using mutual information. Using context-dependent bigrams, with contexts selected by the clustering algorithm, we establish a relative increase in performance of 10% on the attribute error rate on data obtained through the currently operational Swiss automatic train timetable information system.

An adaptive method to learn the compatibility matrix for microcellular systems

In a micro-cellular system, radio propagation prediction may be replaced by statistical analysis of the running system, as far as the distribution of the radio resource between cells is concerned. To this end, interference level and channel usage are sampled and correlated. As a result, the estimation of the compatibility of channels being used at different cell sites can be updated and used for a revised channel assignment. This approach may be seen as a special case of the more general idea of self-optimising networks. The feasibility of the method has been studied by simulations.<<ETX>>