1996/1
Browse
Recent Submissions
Now showing 1 - 5 of 8
- ItemA Two-Level Classification Scheme for CDHMM-Based Discrete-Utterance Recognition(Společnost pro radioelektronické inženýrství, 1996-04) Nouza, J.In the paper a method of speeding up the response of a CDHMM based speech recognition system is introduced. The method, applicable for the recognition of discrete utterances, uses a two-level classification scheme. It consists in a fast match done with simplified models, followed by a final accurate match with a limited number of selected standard models. In this way the recognition time can be reduced by great deal without any significant loss of recognition accuracy. The method has been successfully applied in the design of real-time speech recognition systems operating with small and middle-size vocabularies.
- ItemData Description of a System(Společnost pro radioelektronické inženýrství, 1996-04) Nevriva, P.In this paper, a brief discussion on description of process by memorized data is given. The insight into the problem can offer modified views on optimal control, on data compression at communication systems with respect to information content of message, etc. The idea of process description by memorized data with different information content will be presented here on the classical case study of optimal control: the data based control algorithm (data algorithm, DA) gathers data from the controlled process and derives control signal (control) from data accumulated in the data base. The implementation of the DA on the ideal computer which is not limited by its speed or capacity of memory is expected for simplicity. Accuracy of the data algorithm is then given by a-priori knowledge of the task and by information exchange between the controlled process and the computer.
- ItemA New Glass of Nonlinear Filters: Microstatistic Volterra Filters(Společnost pro radioelektronické inženýrství, 1996-04) Kocur, D.; Drutarovsky, M.; Marchevsky, S.In this paper a new subset of the time-invariant microstatistic filters so-called microstatistic Volterra filters are proposed. This class of nonlinear filters is based on the idea of the conventional microstatistic filter generalization by substituting Wiener filters applied in the conventional microstatistic filter structure by Volterra filters. The advantage of the microstatistic Volterra filters in comparison with the Wiener filters, Volterra filters and conventional microstatistic filters is the fact that in the case of non-Gaussian signal processing the microstatistic Volterra filters can outperform Wiener filters, Volterra filters or conventional microstatistic filters. The validity of this basic property of the microstatistic Volterra filters is verified by a number of computer experiments. The disadvantage of the microstatistic Volterra filters is their relatively high computational complexity.
- ItemInfluence of the Number of the Features with the Neural Network Function(Společnost pro radioelektronické inženýrství, 1996-04) Tuckova, J.; Bores, P.This contribution is devoted to the evaluation of probability of success for classification and recognition techniques. It also depends on a proper selection of input elements-features and their number. Error probability is proportional indirectly on a quantity and quality of information provided to a classifier. It can be affected either by a learning algorithm itself and classification or by an element number on classification correctuers has been verified for Kohonen's map designed for a recognition of Czech digits.
- ItemThe Computation of Forward Scattered Functions for Pruppacher - Pitter Raindrop Forms at 37 GHz(Společnost pro radioelektronické inženýrství, 1996-04) Bazant, L.The amplitude scattered functions were computed for Pruppacher - Pitter raindrop forms at 37 GHz by using the Multiple MultiPole numerical method. Before doing that the measurement of the complex permittivity of rain water had been performed at the same frequency and rain water complex permittivity values had been determined.