DKDijana Kostić

Dijana Kostić

*This is an algorithmically generated author record based on publicly available information.

Dijana Kostić has a total of 10 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Comparative Analysis of the Regression Error in STI Estimation for a Lecture Hall

    by Violeta Stojanović, Zoran Milivojević, Bojan Prlinčević, Dijana Kostić

  • Optimization of the Seventh-order Polynomial Interpolation 1P Kernel in the Time Domain

    by MB University, Zoran N. Milivojevic, Ratko M. Ivkovic, Milan R. Cekic, Academy Of Applied Technical And Preschool Studies, Dijana Z. Kostić, ”Šargan Inženjering” D.O.O.

    This paper presents the optimization of the convolutional, seventh-order polynomial, one-parameter, interpolation kernel. In the first part of the paper, the seventh-order kernel is defined, and, after that, the process of the kernel optimization is described. The optimization criterion was the minimization of the interpolation error e. The optimization involved the selection of the optimal value of the kernel parameter 𝛼, and it was carried out in the time domain. In the second part of this paper, the experiment, which was realized with the aim of determining the precision of interpolation of the third-order, fifth-order, and the seventh-order interpolation kernels, is described. After that a comparative analysis of the interpolation precision is described. As a measure of the interpolation precision, the mean square error (MSE) was used. The results of the experiment are presented graphically and tabularly. Finally, using a comparative analysis, the precision of interpolation with the kernel, whose parameters were optimized in the time domain, in relation to the kernel, whose parameters were optimized in the spectral do-main, was analyzed. Based on the comparative analysis, a recommendation for the optimal parameter for the seventh-order kernel is given.

  • Comparative analysis of LDA, QDA, SVM, algorithms and neural networks for boredom and fear detection in speech

    by Bojan Prlinčević, Zoran Milivojević, Milan Mišić, Dejan Gurešić, Dijana Kostić

    This paper presents an analysis of speech emotion recognition focusing on distinguishing between boredom and fear emotional states. The methodology employs the fundamental frequency trajectory (F₀) and its derivatives to extract characteristics. Emotional states are classified using four machine learning algorithms: Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA), Support Vector Machines (SVM), and Artificial Neural Networks (ANN). The experiment utilizes the Berlin Emotional Speech Database for training and evaluation. Results demonstrate that all classifiers effectively separate boredom and fear emotions, with ANN achieving the highest classification accuracy. Performance analysis are derived using confusion matrix analysis confirm. This study provides an efficient framework for detecting subtle emotional differences in speech signals, particularly between boredom and fear manifestations.