Stanislaw Gmyrek

Stanislaw Gmyrek

Auto-generated profile, based on publicly available information

Stanislaw Gmyrek

Department of Acoustics, Multimedia and Signal Processing, Faculty of Electronics, Photonics and Microsystems Wroclaw University of Science and Technology Wroclaw Poland

Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
11 August 2024 in IET Radar, Sonar & Navigation
Verified
Journal Article
Open access
SGStanislaw Gmyrek
Stanislaw Gmyrek
DSDawid Sysak
Dawid Sysak
PBPawel Biernacki
Pawel Biernacki
GJGrzegorz Jaromi
Grzegorz Jaromi
Drone surveillance with multi‐function radar (MFR) can benefit a lot from the careful radar pulse train organisation into dedicated sub‐RPTs serving particular needs of tasks. In contemporary scenarios of MFR applications, where navigation, search and rescue, and natural environment safeguarding are to be shared, of particular importance is a case where airborne small drone detection radar performs other tasks concurrently. Resource sharing within the radar unit is based on variables and unpredictable circumstances, leading to strains in resources in MFR. Consequently, such constraints limit MFR functioning with optimal drone sensitivity for its own sake. To ensure a high quality of radar services performed quickly, with a low DC power consumption, the task sharing approach by the common MFR platform, non‐periodic bursts emission of radar pulse energy has gained our major attention. At the same time, it was assumed that bursts feature periodic time intervals between radar pulses, which is a favourable circumstance in the control and implementation of signal processing. Each sub‐RPTs is designed to cope with particular detection tasks. One to four linear polarisations (VV, VH, HV, HH) are activated. The focus of the investigation was on the radar emitting a few thousand pulses per second, but only a few 10 pulses per second were expected to be required to detect small airborne drones. The core of the research study was the optimisation process of the scheme operating with 4 to 16 pulse bursts optimised for drone detection with a modest use of resources at MFR. While performing several activities, the X‐band MFR platform called ENAVI was developed in‐house. The proposed non‐periodic burst scheme optimisation approach was validated with in‐field tests with the ENAVI radar model. The in‐air target was a 3‐metre fixed‐wing drone having low RCS and flying at different altitudes with a speed close to 200 km/h. The radar antenna was a 26 dBi low‐profile dual‐polarisation array making it feasible to detect drones remaining a few degrees off the antenna breadboard direction. A very high likelihood of detection of drones from a few kilometres distance was demonstrated within one second. The authors present the effectiveness of the cell‐averaging technique which proved to correctly detect drones within a 7‐km range, and how MFR radar echoes can be used for basic weather surveillance. One vertical VV polarisation was used for drone detection and two VV and VH polarisations for weather observation. Weather radar capabilities were examined against heavy rain clouds imposing a hazard to the safety of drone flights.

25 August 2025 in Archives of Acoustics
Verified
Journal Article
Closed access
PDF file
SGStanislaw Gmyrek
Stanislaw Gmyrek
Urszula LibalUrszula Libal
Urszula Libal
RHRobert Hossa
Robert Hossa
This paper investigates the overfitting problem in vowel classification task for automatic speech recognition (ASR). It utilizes a pitch synchronized human factor cepstral coefficients (PS-HFCC) as the parametrization method, which outperforms traditional methods like HFCC and mel-frequency cepstral coefficients (MFCC) in frame-level classification accuracy. While deep learning models are prevalent in contemporary ASR systems, they often lack explainability, a characteristic of classical classifiers. Therefore, this study examines overfitting phenomenon using a range of classifiers with well-understood properties. Specifically, it analyzes the impact of different training strategies on classifier performance, comparing the susceptibility to overfitting of several widely used classifiers, including the Gaussian mixture model (GMM), a standard approach in speech recognition. The analysis of training strategies considers various data splitting methods: random, speaker-based, and cluster-based. Our analysis of training strategies highlights the crucial role of data splitting methods: while random splitting is commonly used, it can lead to inflated accuracy due to overfitting. We demonstrate that speaker-independent splitting, where the classifier is trained on one set of speakers and tested on a separate, unseen set, is essential for robust evaluation and for accurately assessing generalization to new speakers. Potentially, the resulting insights may inform the future development and training of more reliable ASR systems.

10 March 2023 in Archives of Acoustics
Verified
Journal Article
Closed access
PDF file
PBPawel Biernacki
Pawel Biernacki
SGStanislaw Gmyrek
Stanislaw Gmyrek
WMWladyslaw Magiera
Wladyslaw Magiera
This paper discusses the estimation of flow velocity from a multi-sensor scenario. Different estimation methods were used, which allow the effective measurement of the actual Doppler shift in a noisy environment, such as water with air bubbles, and on this basis the estimation of the flow velocity in the pipe was calculated. Information fusion is proposed for the estimates collected. The proposed approach focuses on the density of the fluid. The proposed method is capable of determining the flow velocity with high accuracy and small variations. Simulation results for plastic and steel (both galvanized and non-galvanized) pipes show the possibility of accurate fluid flow measurements without the need for sensors inside the pipe.

26 September 2023
Verified
Conference Paper
Closed access
SGStanislaw Gmyrek
Stanislaw Gmyrek
RHRobert Hossa
Robert Hossa
RMRyszard Makowski
Ryszard Makowski
The voiced parts of the speech signal are shaped by glottal pulse excitation, the vocal tract, and the speaker's lips. Semantic information contained in speech is shaped mainly by the vocal tract. Unfortunately, the quasiperiodicity of the glottal excitation, in the case of Human Factor Cepstral Coefficients (HFCC) parameterization, is one of the factors affecting the significant scatter of the feature vector values by introducing ripples into the amplitude spectrum. This paper proposes a method to reduce the effect of the quasiperiodicity of the excitation on the feature vector. For this purpose, blind deconvolution was used to determine the vocal tract transfer function estimator and the corrective function of the amplitude spectrum. Then, based on the obtained HFCC parameters, statistical models of individual Polish speech phonemes were developed in the form of mixtures of Gaussian distributions, and the influence of the correction on the quality of classification of speech frames containing Polish vowels was investigated.

1 January 2026 in IET Radar, Sonar & Navigation
Verified
Journal Article
Open access
PKPawel Kabacik
Pawel Kabacik
SGStanislaw Gmyrek
Stanislaw Gmyrek
DSDawid Sysak
Dawid Sysak
ABArkadiusz Byndas
Arkadiusz Byndas
Remotely piloted and autonomous drones, as well as birds and bats, often generate weak radar echoes in many operational cases. The aim of our study, presented in this paper, is to reliably detect such targets, where an operational range of 2–4 km is typically required. Although modern radar techniques are more expensive to implement than current optical methods of airspace analysis, they offer superior performance under adverse weather conditions and day‐and‐night capability compared to optical counterparts. A custom hardware and firmware platform was developed and utilised for real‐time radar data acquisition. To enhance the detection performance of this system, this paper introduces an offline optimisation framework based on Monte Carlo simulations. The framework supports the refinement of radar settings, such as pulse repetition intervals, detection thresholds, and echo averaging parameters, and may improve detection robustness under diverse environmental and meteorological conditions. To validate the effectiveness of the proposed approach, the studies described rely on echoes recorded using the ENAVI radar model operating in the X band (9.5 GHz) with VV polarisation. To substantiate that the significance of our research results extends beyond UAV aviation alone, the last section presents a comparison of real radar echoes from a four‐seat single‐engine aircraft with echoes synthesised using the Monte Carlo method. The proposed method can be used to support offline evaluation of adaptive radar settings under varying ground clutter, such as forests, farmland, mountains, and weather conditions. Simulations carried out evaluate the effectiveness of selected parameters by synthesising thousands of radar echoes, allowing for the rapid verification and adjustment of radar settings within a defined time window.

25 July 2024 in International Journal of Electronics and Telecommunications
Verified
Journal Article
Open access
PDF file
SGStanislaw Gmyrek
Stanislaw Gmyrek
RHRobert Hossa
Robert Hossa
RMRyszard Makowski
Ryszard Makowski
The speech signal can be described by three key elements: the excitation signal, the impulse response of the vocal tract, and a system that represents the impact of speech production through human lips. The primary carrier of semantic content in speech is primarily influenced by the characteristics of the vocal tract. Nonetheless, when it comes to parameterization coefficients, the irregular periodicity of the glottal excitation is a significant factor that leads to notable variations in the values of the feature vectors, resulting in disruptions in the amplitude spectrum with the appearance of ripples. In this study, a method is suggested to mitigate this phenomenon. To achieve this goal, inverse filtering was used to estimate the excitation and transfer functions of the vocal tract. Subsequently, using the derived parameterisation coefficients, statistical models for individual Polish phonemes were established as mixtures of Gaussian distributions. The impact of these corrections on the classification accuracy of Polish vowels was then investigated. The proposed modification of the parameterisation method fulfils the expectations, the scatter of feature vector values was reduced.