prof. Ing. Jan Uhlíř, CSc.

This paper deals with a comparison of k-Means and Bayes classifiers designed for classification of the human body motions classification. The main task of this paper is to compare the performance and the reliability of classifiers. The presented work is a part of research of relations between brain and muscle activity. The sensing of body motions is based on standard DV camcorders system. The procedures have no negative impact to brain activity (the tracking does not affect the measured EEG signals). Presented paper includes the description of observing, discerning and parameterization procedures and the discussion of motion classification. The results of classification accuracy are the part of this paper.

This paper deals with video based parameterization and classification of human body motions. The main task of this work is to develop and verify the procedures for observing of muscle and brain activity. The developed procedures have no negative impact to brain activity (the tracking does not affect the measured EEG signals). The procedures required only standard hardware equipment accessible on neurological laboratories. The body motions are non-contact sensed using a pair of standard DV camcorders. This work includes the description of observing, discerning and parameterization procedures and the discussion of motion classification. The set of classifiers - hierarchical clustering algorithm, recursive clustering algorithm, k-means classifier, Bayes classifier and classifier based on discrimination functions - was developed and implemented. The analysis of the classifiers properties was accomplished in this work. The accuracy of classification was tested for selected

Článek stručně popisuje digitální technologie, které jsou použity při řečové komunikaci. Číslicové zpracování hlasu je aplikováno v telekomunikacích a také při řízení strojů.

The main aim of this paper is the analysis of speech deteriorated by a very rare disease, which induce epileptic seizures in a part of brain responsible for speech production. Speech defects, represented mostly by the combination of missing and mismatched phonemes, are sought and examined in the spectral and time domain. An algorithm, proposed in this paper, is based on Hidden Markov Models (HMMs) and it is most suitable for the speech recognition tasks. The algorithm is able to analyze in both time and spectral domains simultaneously; in the spectral domain as a log-likelihood score and in the time domain as a forced time alignment of the HMMs. The suggested algorithm works properly in the time domain. The results for the spectral domain are not credible, because the algorithm have to be tested on more data (not available at the time of paper preparation).

Monografie přináší ucelenou informaci o metodách číslicového zpracování signálů lidské řeči, a to v oblastech přenosu, syntézy a rozpoznávání. Poslouží jako studijní literatura k předmětu Digitální zpracování signálu řeči, který je zahrnut do magisterského programu strukturovaného studia i předmětu Fonetické signály a jejich kódování, který je ve studijním plánu doktorského programu. Jejím posláním je nejen seznámit širší veřejnost s touto vědní disciplínou všeobecně, ale publikovat i některé vědecké výsledky získané při výzkumu v laboratoři katedry teorie obvodů na Fakultě elektrotechnické ČVUT.

This paper deals with the classification of three dimensional thumb motions. The thumb motion is parameterized tracing a special mark on the thumb, the coordinates of thumb rajectory are computed during the parameterization. The discrimination functions used for the classification of thumb motion are derived in this paper. The discrimination functions are based on statistical data analysis. The functions are derived from the Bayes theorem. The normal distribution of motion parameters is assumed. The parameters of discrimination functions are estimated using the Maximum Likelihood Estimation. The presented lassification is used in biomedical engineering for research on correlations between a human brain function and a muscle activity.

Achieving reliable performance in speech recognition is the car for mobile telephony application has been studying intensively for more than one decade. This paper addresses effects of mismatched conditions and their minimization with respect to the performance of speaker-independent isolated. word recognition in the car-noise environment without consideration of Lombard effect. This study is primarily intended to study the dependence of the recognition rate on the SNR of an input signal without and with noise enhancement preprocessing, especially to find conditions under that the modified spectral subtraction call be effectively used for the speech recognition ill a real non-stationary car-noise environment. If as the worst recognition rate. is admitted e.g. 80%, then tile use of spectral subtraction methods enables to use wider interval of input SNRs: for the trainig: made oil a clean speech this interval is (40,6)(1) dB; for the training made on a noisy speech this interval is (40,-2) dB; for tile training performed oil an enhanced speech this interval is (40,-8) dB. The third case gives tile widest interval of SNRs in which a recogniser (with tile final recognition rate ill tile interval of (100,80)%) can be used.