BME MDA - a Műegyetem Digitális Archivuma
Not a member yet
47877 research outputs found
Sort by
Controlling Intelligent Prostheses Using ETSI MEC
No full textCurrently, algorithms required to operate intelligent prostheses are usually executed locally, using the embedded system integrated into prostheses. However, the limited computational capabilities of such devices usually constrain the complexity and thus the accuracy of the algorithms that determine the intention of the user, resulting in a sub-optimal user experience. Modern networking technologies in combination with edge cloud solutions offer significantly more raw computational power (in comparison with embedded solutions). With predictable latency and low overhead, more robust and accurate algorithms could be applied for processing data collected by sensors, thus providing more accurate control, at the cost of the latency introduced by the network. The goal of my research project is to determine whether 5G networks and edge cloud solutions are viable as the background for offloading parts of a control system to edge cloud applications and address the challenges of offloading control in time-critical settings with a relatively high-frequency input such as electromyography (EMG) sensors. This paper describes a framework for offloading near real-time computations using an IP network along with a set of tools for benchmarking networks and the protocol itself in different network conditions. The control of intelligent prostheses is an application the framework is specifically tailored to. EMG data is collected and used to determine the intended action, with the accuracy depending on the complexity of a classifier algorithm. Other applications related to control systems with similar requirements were also kept in mind during the design phase. The capabilities of different networking technologies in combination with the framework are shown through the results of various benchmarks and experiments
Célzott szakmai interjúk mint módszertani eljárás a történelemtanítás helyzetének tükrében
No full textA tanulmány a célzott szakmai interjúkkal mint módszertani eljárással foglalkozik a történelemtanítás helyzetének tükrében. A történelemtanítás hatékonysága kiemelten fontos a további történelemdidaktikai kutatások szempontjából, ezért alapvető, hogy a jelenlegi történelemtanítási helyzettel tisztában legyünk. A szakmai interjúk módszertana hatékony eszköz lehet ennek kutatására. Jelen kutatás egyik célja, hogy bemutasson egy olyan módszertani eljárást, amely alkalmazható a pedagógia és tantárgyak helyzetének kutatására. Emellett a másik célja a feltérképezés a történelem tantárgyról, azon belül a forráselemzéssel kapcsolatos jelenlegi nehézségekről. A kutatás során 12 félig strukturált interjú készült, amelyek elemzése a módszertani eljárás alkalmazhatóságát és hatékonyságát is vizsgálja. Az eredmények arra utalnak, hogy a történelemtanítás során jelentkező nehézségek az iskolatípustól és az oktatási szinttől függően változnak. Ezek a problémák elsősorban a tanulók képességeiben mutatkoznak meg, különösen a forráselemzés területén, illetve egyéb tényezők is nehézséget okoznak a történelemtanárok mindennapi munkájában. A szakmai interjúk alkalmazásával szélesebb és mélyebb kép alkotható a (történelem)tanításról. Alkalmazása lehetőséget biztosít a kutatók számára, hogy képet kapjanak többek között a pedagógusok mindennapi gyakorlatáról és feltételezett nehézségeiről. Eredményként a kutatás a szakmai interjúk módszertani eljárását pozitívan értékeli, s alkalmazását tovább hirdeti más kutatók számára
A Hybrid Algorithm for Robust Pitch Estimation in Emotional Speech Synthesis
No full textEmotional intelligence in synthetic speech remains a critical challenge in human-machine interaction, despite significant advances in speech synthesis naturalness and intelligibility. Current systems struggle to accurately capture the nuanced emotional expressions characteristic of human speech, including rapid pitch transitions, wide frequency variations, and irregular vibrato patterns. While pitch estimation algorithms like PESTO and FCPE have proven effective for standard speech, their performance on emotional content remains largely unexplored. We present ESCAPE (Emotion Self-Supervised ContextAware Pitch Estimation), a novel algorithm specifically designed for emotional speech processing. ESCAPE synthesizes PESTO's precise frequency variation handling with FCPE's context-aware processing through a hybrid architecture that achieves robust pitch tracking in expressive vocal content. Our approach maintains computational efficiency while excelling at capturing complex acoustic patterns unique to emotional utterances. This paper provides the first comprehensive evaluation of PESTO and FCPE on emotional speech datasets and introduces ESCAPE as a transformative solution for pitch estimation in emotionally expressive speech synthesis. Our results demonstrate significant progress toward bridging the gap between human-like emotional expression and machine-generated speech, marking an important advancement in emotional speech synthesis technology
Optimizing Cardiac MRI Segmentation: An Ensemble Approach with U-Net Variants
No full textSegmentation of cardiac magnetic resonance images is a critical task in medical imaging, particularly to delineate the left and right ventricles and the myocardium. This study aims to improve segmentation performance using an ensemble approach with variants of the U-Net architecture, a widely adopted deep learning model for image segmentation. Multiple segmentation models were trained and optimized, and their outputs were combined using threshold-based binary conversion. Two ensemble strategies were evaluated: (1) Averaging, where the mean value of the binary masks at each pixel location was calculated to smooth discrepancies among model predictions, and (2) Voting, where majority voting determined the final pixel classification. The proposed ensemble approach demonstrates robustness to individual model errors and improves segmentation consistency
Effects of Noisy Occupancy Data on an Auction-based Intelligent Parking Assignment
No full textSmartphones and cloud services can provide sophisticated parking assignment in modern intelligent cities. These solutions aim to guide drivers to vacant parking lots near their destination, reducing the necessary cruising for parking. Hence, they can smoothen the traffic flow and mitigate harmful emissions. Moreover, auction-based assignment can also dynamically optimize the actual parking prices, benefiting drivers, and parking lot operators.
To operate such a system, we shall know the actual occupancy of the supervised parking lots. This data can come from various sources, e.g., crowdsourcing, parking lot operators, or third-party data providers. Sensing and fusing these records might lead to inaccurate input for the assignment method. In this paper, we analyze the impact of such noise on the performance of an auction-based parking lot assignment system. The results indicate that accurate information is crucial for perfect operation, but current state-of-the-art solutions provide sufficient input to benefit from the system
