Đorđe Damnjanović
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Đorđe Damnjanović has a total of 27 co-authors. Here are the top co-authors by number of shared publications:
- Dejan Vujičić
Faculty of Technical Sciences Čačak, University of Kragujevac, 32102 Čačak, Serbia
7 shared publications - Marina Milosevic
Department of Computer and Software Engineering, University of Kragujevac, Faculty of Technical Sciences Cacak, Svetog Save 65, Cacak, Serbia
6 shared publications - Dijana Stojić
University of Kragujevac,Kragujevac,Faculty of Technical Sciences,Serbia
4 shared publications - Dijana Stojić3 shared publications
- Branko R. Marković
Department of Computer and Software Engineering, Faculty of Technical Sciences Čačak, University of Kragujevac, Čačak, Serbia
2 shared publications Dejan Vujičić
University of Kragujevac,Kragujevac,Faculty of Technical Sciences,Serbia
2 shared publications
Latest Publications
- 1 January 2026
- Published by University Library in Kragujevac
- p. 210-215
SerbNumAO 1.0: Design, Development, and Applications of a Serbian Speech Database of Numbers and Arithmetic Operators
This paper presents the design and development of a Serbian speech database, primarily consisting of spoken numbers and arithmetic operators. The vocabulary comprises 54 Serbian words. The database includes recordings from 50 speakers, each of whom uttered the complete vocabulary twice using a normal speaking style. The primary objective of this database is to support human–machine communication applications. To verify its applicability for speech recognition tasks, a subset of the SerbNumAO 1.0 database was used in an automatic speech recognition experiment. Mel-Frequency Cepstral Coefficients (MFCCs) were employed as front-end of ASR (Automatic Speech Recognition) system, while a Convolutional Neural Network (CNN) served as the back-end classifier. The proposed method achieved an average Word Recognition Rate (WRR) of up to 95%.
- 1 January 2026
- Published by University Library in Kragujevac
- p. 210-215
SerbNumAO 1.0: Design, Development, and Applications of a Serbian Speech Database of Numbers and Arithmetic Operators
This paper presents the design and development of a Serbian speech database, primarily consisting of spoken numbers and arithmetic operators. The vocabulary comprises 54 Serbian words. The database includes recordings from 50 speakers, each of whom uttered the complete vocabulary twice using a normal speaking style. The primary objective of this database is to support human–machine communication applications. To verify its applicability for speech recognition tasks, a subset of the SerbNumAO 1.0 database was used in an automatic speech recognition experiment. Mel-Frequency Cepstral Coefficients (MFCCs) were employed as front-end of ASR (Automatic Speech Recognition) system, while a Convolutional Neural Network (CNN) served as the back-end classifier. The proposed method achieved an average Word Recognition Rate (WRR) of up to 95%.
- 1 January 2026
- Published by University Library in Kragujevac
- p. 120-127
Usability Evaluation of the Electronic Gradebook Interface from the Teachers’ Point of View
The interface of information systems used in secondary schools strongly affects how efficiently teaching and administration are carried out. As the main users of these systems, teachers play an important role in assessing their usefulness and functionality. This paper examines how teachers perceive the interface of the electronic gradebook (esDnevnik), the system that is mandatory in all schools in Serbia and represents the teacher’s principal everyday tool. An empirical usability study was conducted with thirty-one teachers of a vocational secondary school in Čačak. Each participant completed a set of predefined tasks reflecting routine work in the system, after which they filled in a questionnaire about their experience. The sessions were recorded and analysed with the Morae software package, which measured task success, the number of mouse clicks and the time needed to complete each task. The results show that teachers value the intuitiveness and ease of use of the interface, but that there are notable problems with system speed, the visibility of notifications, and technical support. Although objective usability testing revealed differences in the way class teachers and subject teachers perform particular tasks, no statistically significant differences were found in their subjective evaluation of the system usability. Based on the findings, recommendations are proposed for improving the technical infrastructure, user training and communication between administrators and teachers.
- 1 January 2026
- Published by University Library in Kragujevac
- p. 120-127
Usability Evaluation of the Electronic Gradebook Interface from the Teachers’ Point of View
The interface of information systems used in secondary schools strongly affects how efficiently teaching and administration are carried out. As the main users of these systems, teachers play an important role in assessing their usefulness and functionality. This paper examines how teachers perceive the interface of the electronic gradebook (esDnevnik), the system that is mandatory in all schools in Serbia and represents the teacher’s principal everyday tool. An empirical usability study was conducted with thirty-one teachers of a vocational secondary school in Čačak. Each participant completed a set of predefined tasks reflecting routine work in the system, after which they filled in a questionnaire about their experience. The sessions were recorded and analysed with the Morae software package, which measured task success, the number of mouse clicks and the time needed to complete each task. The results show that teachers value the intuitiveness and ease of use of the interface, but that there are notable problems with system speed, the visibility of notifications, and technical support. Although objective usability testing revealed differences in the way class teachers and subject teachers perform particular tasks, no statistically significant differences were found in their subjective evaluation of the system usability. Based on the findings, recommendations are proposed for improving the technical infrastructure, user training and communication between administrators and teachers.
- Open Access
- 21 November 2025
- Published by University of Business Engineering and Management Banja Luka
- Vol. 14 (2) , 209-215
IMPLEMENTATION OF WAVELET SCALOGRAMS FOR AUDIO SIGNAL ANALYSIS
In the past few decades, wavelets have found an important place in many fields due to their nature and advantages compared to other algorithms. When the focus is on signal processing, wavelets can be used as an algorithm for de-noising, frequency analysis, feature analysis, etc. The spectrogram is one of the most common visual representations of the signal in the frequency domain when the Fourier transformation is used. Wavelets have their algorithm for signal representation in the frequency domain, and it’s called scalogram due to the nature of the scaling properties of the wavelet function. This paper explores the application of the Continuous Wavelet Transform (CWT) for generating and analyzing scalograms. Audio signals from different direct current (DC) motor sounds are used as test data to demonstrate the effectiveness of this approach. All implementations and analyses are carried out using MATLAB software. The results highlight the advantages of wavelet-based analysis in capturing time-frequency characteristics that may not be easily observed with traditional Fourier-based methods.
- 21 November 2025
- Published by University of Business Engineering and Management Banja Luka
- Vol. 14 (2) , 209-215
Open AccessIMPLEMENTATION OF WAVELET SCALOGRAMS FOR AUDIO SIGNAL ANALYSIS
In the past few decades, wavelets have found an important place in many fields due to their nature and advantages compared to other algorithms. When the focus is on signal processing, wavelets can be used as an algorithm for de-noising, frequency analysis, feature analysis, etc. The spectrogram is one of the most common visual representations of the signal in the frequency domain when the Fourier transformation is used. Wavelets have their algorithm for signal representation in the frequency domain, and it’s called scalogram due to the nature of the scaling properties of the wavelet function. This paper explores the application of the Continuous Wavelet Transform (CWT) for generating and analyzing scalograms. Audio signals from different direct current (DC) motor sounds are used as test data to demonstrate the effectiveness of this approach. All implementations and analyses are carried out using MATLAB software. The results highlight the advantages of wavelet-based analysis in capturing time-frequency characteristics that may not be easily observed with traditional Fourier-based methods.
- Open Access
- 21 November 2025
- Published by University of Business Engineering and Management Banja Luka
- Vol. 14 (2) , 201-208
PERFORMANCE COMPARISON OF VARIOUS DATABASE TYPES IN REACT NATIVE APPLICATIONS
In the modern digital age, mobile devices have become an indispensable part of everyday life. With the growing volume of data we handle daily, the speed of data loading and access becomes a crucial factor for the efficient operation of mobile applications. Large datasets are becoming increasingly common, such as, for example, the German collocation dictionary. Given that mobile devices use different operating systems, such as Android and iOS, there is an increasing need for tools that enable the development of cross-platform applications from a single codebase. React Native, as a popular open-source framework, provides exactly this solution. This paper tested the loading of data from the dictionary using React Native applications. This data loading was performed in four ways: using AsyncStorage, SQLite, TinyBase, and Realm databases. A different number of rows was loaded, and the loading speed was compared.
- 21 November 2025
- Published by University of Business Engineering and Management Banja Luka
- Vol. 14 (2) , 201-208
Open AccessPERFORMANCE COMPARISON OF VARIOUS DATABASE TYPES IN REACT NATIVE APPLICATIONS
In the modern digital age, mobile devices have become an indispensable part of everyday life. With the growing volume of data we handle daily, the speed of data loading and access becomes a crucial factor for the efficient operation of mobile applications. Large datasets are becoming increasingly common, such as, for example, the German collocation dictionary. Given that mobile devices use different operating systems, such as Android and iOS, there is an increasing need for tools that enable the development of cross-platform applications from a single codebase. React Native, as a popular open-source framework, provides exactly this solution. This paper tested the loading of data from the dictionary using React Native applications. This data loading was performed in four ways: using AsyncStorage, SQLite, TinyBase, and Realm databases. A different number of rows was loaded, and the loading speed was compared.
- Open Access
- 21 November 2025
- Published by University of Business Engineering and Management Banja Luka
- Vol. 14 (2) , 207-213
IMPLEMENTATION OF WAVELET SCALOGRAMS FOR AUDIO SIGNAL ANALYSIS
In the past few decades, wavelets have found an important place in many fields due to their nature and advantages compared to other algorithms. When the focus is on signal processing, wavelets can be used as an algorithm for de-noising, frequency analysis, feature analysis, etc. The spectrogram is one of the most common visual representations of the signal in the frequency domain when the Fourier transformation is used. Wavelets have their algorithm for signal representation in the frequency domain, and it’s called scalogram due to the nature of the scaling properties of the wavelet function. This paper explores the application of the Continuous Wavelet Transform (CWT) for generating and analyzing scalograms. Audio signals from different direct current (DC) motor sounds are used as test data to demonstrate the effectiveness of this approach. All implementations and analyses are carried out using MATLAB software. The results highlight the advantages of wavelet-based analysis in capturing time-frequency characteristics that may not be easily observed with traditional Fourier-based methods.
- 21 November 2025
- Published by University of Business Engineering and Management Banja Luka
- Vol. 14 (2) , 207-213
Open AccessIMPLEMENTATION OF WAVELET SCALOGRAMS FOR AUDIO SIGNAL ANALYSIS
In the past few decades, wavelets have found an important place in many fields due to their nature and advantages compared to other algorithms. When the focus is on signal processing, wavelets can be used as an algorithm for de-noising, frequency analysis, feature analysis, etc. The spectrogram is one of the most common visual representations of the signal in the frequency domain when the Fourier transformation is used. Wavelets have their algorithm for signal representation in the frequency domain, and it’s called scalogram due to the nature of the scaling properties of the wavelet function. This paper explores the application of the Continuous Wavelet Transform (CWT) for generating and analyzing scalograms. Audio signals from different direct current (DC) motor sounds are used as test data to demonstrate the effectiveness of this approach. All implementations and analyses are carried out using MATLAB software. The results highlight the advantages of wavelet-based analysis in capturing time-frequency characteristics that may not be easily observed with traditional Fourier-based methods.
- Open Access
- 21 November 2025
- Published by University of Business Engineering and Management Banja Luka
- Vol. 14 (2) , 199-206
PERFORMANCE COMPARISON OF VARIOUS DATABASE TYPES IN REACT NATIVE APPLICATIONS
In the modern digital age, mobile devices have become an indispensable part of everyday life. With the growing volume of data we handle daily, the speed of data loading and access becomes a crucial factor for the efficient operation of mobile applications. Large datasets are becoming increasingly common, such as, for example, the German collocation dictionary. Given that mobile devices use different operating systems, such as Android and iOS, there is an increasing need for tools that enable the development of cross-platform applications from a single codebase. React Native, as a popular open-source framework, provides exactly this solution. This paper tested the loading of data from the dictionary using React Native applications. This data loading was performed in four ways: using AsyncStorage, SQLite, TinyBase, and Realm databases. A different number of rows was loaded, and the loading speed was compared.
- 21 November 2025
- Published by University of Business Engineering and Management Banja Luka
- Vol. 14 (2) , 199-206
Open AccessPERFORMANCE COMPARISON OF VARIOUS DATABASE TYPES IN REACT NATIVE APPLICATIONS
In the modern digital age, mobile devices have become an indispensable part of everyday life. With the growing volume of data we handle daily, the speed of data loading and access becomes a crucial factor for the efficient operation of mobile applications. Large datasets are becoming increasingly common, such as, for example, the German collocation dictionary. Given that mobile devices use different operating systems, such as Android and iOS, there is an increasing need for tools that enable the development of cross-platform applications from a single codebase. React Native, as a popular open-source framework, provides exactly this solution. This paper tested the loading of data from the dictionary using React Native applications. This data loading was performed in four ways: using AsyncStorage, SQLite, TinyBase, and Realm databases. A different number of rows was loaded, and the loading speed was compared.
- Open Access
- 24 September 2025
- Published by MDPI AG in Electronics
- Vol. 14 (19) , 3793
Deep Learning System for Speech Command Recognition
We present a deep learning model for the recognition of speech commands in the English language. The dataset is based on the Google Speech Commands Dataset by Warden P., version 0.01, and it consists of ten distinct commands (“left”, “right”, “go”, “stop”, “up”, “down”, “on”, “off”, “yes”, and “no”) along with additional “silence” and “unknown” classes. The dataset is split in a speaker-independent manner, with 70% of speakers assigned to the training set and 15% to the test set and validation set. All audio clips are sampled at 16 kHz, with a total of 46 146 clips. Audio files are converted into Mel spectrogram representations, which are then used as input to a deep learning model composed of a four-layer convolutional neural network followed by two fully connected layers. The model employs Rectified Linear Unit (ReLU) activation, the Adam optimizer, and dropout regularization to improve generalization. The achieved testing accuracy is 96.05%. Micro- and macro-averaged precision, recall, and F1-score of 95% are reported to reflect class-wise performance, and a confusion matrix is also provided. The proposed model has been deployed on a Raspberry Pi 5 as a Fog computing device for real-time speech recognition applications.
- 24 September 2025
- Published by MDPI AG in Electronics
- Vol. 14 (19) , 3793
Open AccessDeep Learning System for Speech Command Recognition
We present a deep learning model for the recognition of speech commands in the English language. The dataset is based on the Google Speech Commands Dataset by Warden P., version 0.01, and it consists of ten distinct commands (“left”, “right”, “go”, “stop”, “up”, “down”, “on”, “off”, “yes”, and “no”) along with additional “silence” and “unknown” classes. The dataset is split in a speaker-independent manner, with 70% of speakers assigned to the training set and 15% to the test set and validation set. All audio clips are sampled at 16 kHz, with a total of 46 146 clips. Audio files are converted into Mel spectrogram representations, which are then used as input to a deep learning model composed of a four-layer convolutional neural network followed by two fully connected layers. The model employs Rectified Linear Unit (ReLU) activation, the Adam optimizer, and dropout regularization to improve generalization. The achieved testing accuracy is 96.05%. Micro- and macro-averaged precision, recall, and F1-score of 95% are reported to reflect class-wise performance, and a confusion matrix is also provided. The proposed model has been deployed on a Raspberry Pi 5 as a Fog computing device for real-time speech recognition applications.
- 1 January 2025
- Published by Akademska Misao
- p. 103-107
Izdvajanje obeležja na bazi spektograma za klasifikaciju radarskih ciljeva
U ovom radu analizirani su metodi za izdvajanje obeležja radarskih ciljeva (vozila, osobe, grupe) zasnovanih na teksturi spektrograma. Korišćeni su parametri: entropija, moment trećeg reda raspodele energije i usmerenost spektrograma. Istraživanje je pokazalo da izbor parametara za izračunavanje spektrograma (vrsta prozora, dužina prozora, preklapanje i broj tačaka Furijeove transformacije) značajno utiče na separabilnost klasa. Najbolji rezultati separabilnosti ostvareni su za pravougaoni prozor dužine 512 odmeraka. Obeležja zasnovana na teksturi spektrograma pokazala su potencijal za automatsku klasifikaciju sa tačnošću većom od 90%.
- 1 January 2025
- Published by Akademska Misao
- p. 103-107
Izdvajanje obeležja na bazi spektograma za klasifikaciju radarskih ciljeva
U ovom radu analizirani su metodi za izdvajanje obeležja radarskih ciljeva (vozila, osobe, grupe) zasnovanih na teksturi spektrograma. Korišćeni su parametri: entropija, moment trećeg reda raspodele energije i usmerenost spektrograma. Istraživanje je pokazalo da izbor parametara za izračunavanje spektrograma (vrsta prozora, dužina prozora, preklapanje i broj tačaka Furijeove transformacije) značajno utiče na separabilnost klasa. Najbolji rezultati separabilnosti ostvareni su za pravougaoni prozor dužine 512 odmeraka. Obeležja zasnovana na teksturi spektrograma pokazala su potencijal za automatsku klasifikaciju sa tačnošću većom od 90%.
- Open Access
- 1 January 2025
- Published by Centre for Evaluation in Education and Science (CEON/CEES) in Engineering Today
- No. 00,p. 45
Wavelet-based feature analysis of vehicle vibration signals for comfort assessment
This study explores the use of wavelet-based analysis for evaluating vehicle ride comfort through vibration signal processing. Using a custom-developed Android application, three-axis accelerometer and GPS data were collected under real driving conditions to identify three categories of passenger experience: comfortable, vibrationally uncomfortable, and oscillatory uncomfortable. The signals were decomposed using the Daubechies wavelet (Db2) up to level 8, and feature extraction was conducted using detail coefficients. Significant differences between signal types were observed through time and frequency domain analyses, with enhanced discrimination achieved via wavelet decomposition and RMS-based quantification. The results highlight that the vibrationally uncomfortable signals exhibit the highest coefficient values, clearly differentiating them from the other two categories. The findings confirm that wavelet-based methods provide improved insights into vibration signal characteristics, enabling more precise classification of discomfort. This work establishes a foundation for real-time comfort monitoring and future applications in transportation safety and healthcare, particularly in the context of patient transport.
- 1 January 2025
- Published by Centre for Evaluation in Education and Science (CEON/CEES) in Engineering Today
- No. 00,p. 45
Open AccessWavelet-based feature analysis of vehicle vibration signals for comfort assessment
This study explores the use of wavelet-based analysis for evaluating vehicle ride comfort through vibration signal processing. Using a custom-developed Android application, three-axis accelerometer and GPS data were collected under real driving conditions to identify three categories of passenger experience: comfortable, vibrationally uncomfortable, and oscillatory uncomfortable. The signals were decomposed using the Daubechies wavelet (Db2) up to level 8, and feature extraction was conducted using detail coefficients. Significant differences between signal types were observed through time and frequency domain analyses, with enhanced discrimination achieved via wavelet decomposition and RMS-based quantification. The results highlight that the vibrationally uncomfortable signals exhibit the highest coefficient values, clearly differentiating them from the other two categories. The findings confirm that wavelet-based methods provide improved insights into vibration signal characteristics, enabling more precise classification of discomfort. This work establishes a foundation for real-time comfort monitoring and future applications in transportation safety and healthcare, particularly in the context of patient transport.
- 31 July 2024
- Published by Akademska Misao
- p. 9-13
Experiments in Whispered Speech Recognition Based on Wavelet Transformation
In this paper the results of normal and whispered speech recognition based on DWT (Discrete Wavelet Transformation) are presented. The feature vectors are obtained based on Daubechies sub-band energy. The experiments are performed using a part of the Whi-Spe database (one female and one male speaker). A back-end of the ASR system is based on DTW (Dynamic Time Warping) algorithm. The following scenarios are analyzed: normal/normal, whisper/whisper, normal/whisper and whisper/normal in the speaker dependent mode. The results confirmed the usefulness of Wavelet transformation in speech recognition.
- 31 July 2024
- Published by Akademska Misao
- p. 9-13
Experiments in Whispered Speech Recognition Based on Wavelet Transformation
In this paper the results of normal and whispered speech recognition based on DWT (Discrete Wavelet Transformation) are presented. The feature vectors are obtained based on Daubechies sub-band energy. The experiments are performed using a part of the Whi-Spe database (one female and one male speaker). A back-end of the ASR system is based on DTW (Dynamic Time Warping) algorithm. The following scenarios are analyzed: normal/normal, whisper/whisper, normal/whisper and whisper/normal in the speaker dependent mode. The results confirmed the usefulness of Wavelet transformation in speech recognition.