18069 research outputs found
Sort by
Analysis of Individual Speaker Features on Group-level Emotion Recognition from Speech
Get PDFGroup Emotion Recognition (GER) ist von entscheidender Bedeutung sowohl f¨ur das Verst¨andnis sozialer Dynamiken als auch f¨ur das zwischenmenschliche Verhalten und f¨ur die Verbesserung von Mensch-Computer-Interaktionen. Group-Emotions umfassen die von den einzelnen Mitgliedern empfundenen Emotionen und die Kontextfaktoren der Gruppe, die den emotionalen Zustand der Gruppe beeinflussen. Systeme f¨ur GER, die auf Sprachsignalen basieren, sollten sich daher auf Informationen st¨utzen, die sowohl aus dem Gruppenkontext als auch aus dem individuellen Kontext stammen. Die Verwendung von Features, die aus Bildern von Einzelpersonen extrahiert wurden, in Kombination mit gruppenbasierten Features hat sich als wirksam f¨ur die Verbesserung der GER erwiesen. Daher l¨asst sich annehmen, dass die Einbeziehung individueller Informationen von Sprechenden zur Verbesserung der speech-based GER f¨uhren kann. In dieser Arbeit soll untersucht werden, wie verschiedene Arten von akustischen Features (cepstral, spectral, prosodic, temporal und Sound-Quality-Features) der einzelnen Sprechenden zur Emotionserkennung der Gruppe beitragen. Um dies zu erreichen, wird ein vortrainiertes Modell zur Speech-Separation verwendet. Dadurch werden die einzelnen Stimmen der Speech-Mixture isoliert. Anschließend werden Features aus der Speech-Mixture (Mixture-Features) und der Sprache der einzelnen Sprechenden (individual Features) extrahiert. Es werden verschiedene Experimente mit Support Vector Machines (SVMs) und Fully-Connected Neural Networks (FCNNs) f¨ur speech-based GER durchgef¨uhrt. Die Modelle werden f¨ur die Erkennung von drei Group-Emotiuon- Classes, namentlich Positiv, Neutral, und Negativ trainiert und auf speaker-dependant (bekannte Sprechende) und speaker-indipendent (unbekannte Sprechende) Daten getestet. Die verwendeten Sprachdaten bestehen aus Speech-Mixtures, die aus Videos der VGAF-Datenbank gewonnen wurden. Die SVM- und FCNN-Modelle werden separat auf verschiedenen Feature-Sets trainiert, die Mixture- und individual Features kombinieren. Die Ergebnisse zeigen eine signifikante Verbesserung der Leistung des FCNN-Modells in speaker-indipendent Test-Szenarien, wenn das Modell eine Kombination aus spectral-individual Features und Mixture-Features als Trainingsdaten verwendet (Makro-F1-Score = 65.56%), verglichen mit der Verwendung von ausschließlich Mixture-Features (Makro-F1-Score = 53.48%). Sound-Quality-, temporal und cepstral Features der einzelnen Sprechenden zeigen ebenfalls Verbesserungen bei den GERWerten, die von den Modellen erreicht werden, wenn sie in Kombination mit den Mixture-Features f¨ur das Training verwendet werden.Group Emotion Recognition (GER) is crucial for understanding social dynamics and inter-human behavior, enhancing human-computer interactions. Group emotion encompasses the emotions felt by the individual members and group context factors that shape the emotional state of the group. Employing features extracted from images of the individuals in combination with group-based features has been proven to be effective for improving GER. Consequently, there is a potential for involving individual speaker information to improve speech-based GER. This work aims to analyze how different types of acoustic features (cepstral, spectral, prosodic, temporal, and sound quality features) from the individual speakers contribute to the emotion recognition of the group. To achieve this, a pre-trained model for speech separation is employed for isolating the speech of each speaker in the mixture. Then, features are extracted from the speech mixtures (mixture features) and the individuals’ speech (individual features). Various experiments are conducted using Support Vector Machines (SVMs) and Fully-Connected Neural Networks (FCNNs) for speech-based GER. The models are trained for recognition of three group emotion classes, namely Positive, Neutral, and Negative. They are tested in speaker-dependent, i.e. known speakers, as well as speaker-independent, i.e. unknown speakers, scenarios. The speech data employed consists of speech mixtures obtained from videos of the VGAF database. The SVM and FCNN models are trained separately on different feature sets that combine mixture and individual features. The results show a significant improvement in the performance of the FCNN model in speaker-independent scenarios when the model uses a combination of spectral-individual features and mixture features as training input (macro F1-score = 65.56%), compared to using only mixture features (macro F1-score = 53.48%). Sound quality, temporal, and cepstral features from the individual speakers also demonstrate improvements in the GER scores achieved by the models when used in combination with the mixture features for training
Unleashing the power of intelligence : revolutionizing malaria outbreak preparedness with an advanced warning system in Benin, West Africa
No full textBackground: Malaria is a significant vector-borne disease that exhibits high sensitivity to climatic variations within the West African region. In Benin, the effective prevention and mitigation of malaria pose considerable challenges, primarily due to the prevailing conditions of poverty and environmental adversities. This study endeavours to devise an advanced system for early detection and warning of malaria outbreaks in the northern part of Benin, employing monthly time series data pertaining to climatic variables. Methods: Monthly climate data were sourced from Meteorological Agency of Benin (METEO-Benin), alongside malaria incidence data procured from the database of the Benin Ministry of Health, that covered the timeframe of 2009–2021. To ascertain the influence of climatic variables on malaria incidence, principal component analysis was applied. Subsequently, an intelligent model for forecasting malaria outbreaks was developed using support vector machine (SVM) algorithm. The developed model for malaria outbreaks was then employed to establish an intelligent system for warning and forecasting malaria incidence on a monthly basis, utilising the Meteostat platform, an online weather data service provider, in conjunction with the Streamlit framework. This application exhibits responsiveness and compatibility across all web browsers. Results: Relative humidity and maximal temperature significantly influence malaria incidence in the northern region of Benin. SVM regression algorithm forecasts 80% prediction rate for malaria incidence. Consequently, the intelligent malaria outbreak warning system was successfully devised, enabling the automatic and manual prediction of monthly malaria incidence rates within the districts of northern Benin. Conclusions: This system serves as a valuable tool for stakeholders and policymakers, facilitating proactive measures to curtail malaria transmission in Benin.PeerReviewe
Sustainable science or business as usual? Exploring awareness and actions in German University Laboratories
No full textAs sustainability research covers many disciplines, it is important to understand how sustainable research itself is. Previous studies have shown that laboratories, a core infrastructure of research, consume a large number of resources. However, overall understanding and identification of actions that can improve the current situation is rather unclear due to a large data gap on the opinions and attitudes of staff and students within institutions. This study explored sustainable actions among laboratory staff and students in a sample of German universities, in the categories of materials, devices, procurement, waste disposal, and awareness. For this purpose, a mixed-method approach was applied, consisting of a quantitative survey answered by 81 laboratories, and a qualitative interview conducted within 9 different universities. One of the key findings is that the inclusion of more sustainable laboratory activities (e.g. efficient waste and equipment management) was mostly undertaken by the laboratory staff themselves, which led to many bottom-up approaches. However, the necessary framework from the administration and institutional support is often missing or is rather limited, which hinders progress in existing initiatives and to take advantage of their great potential. To achieve effective change, it is essential to engage all stakeholders in the institution, laboratory personnel, and institutional administrators. The findings provide valuable insights into the landscape of German university laboratories and their actions and highlight potential areas for structural intervention to make the laboratory operations in universities more sustainable.PeerReviewe
Effect of the GHG quota for the electromobility market ramp-Up in Northern Germany
No full textPeerReviewe
Caring community as a model for care and support for people in need of care : a scoping review
No full textEinleitung: Das Konzept der Sorgenden Gemeinschaft beschreibt einen Ansatz, bei dem nicht nur professionelle Pflegekräfte, sondern Familien, Nachbar:innen, Ehrenamtliche, lokale Organisationen und Kommunen gemeinsam Verantwortung für ältere oder unterstützungsbedürftige Menschen übernehmen. Digitale Technologien können dabei die Vernetzung und Kommunikation erleichtern sowie Zusammenarbeit über Quartiersgrenzen hinweg ermöglichen. Methode: Zur Auswertung der Forschungsliteratur wurde ein Scoping Review nach dem Joanna Briggs Institute durchgeführt. Die systematische Suche erfolgte in vier wissenschaftlichen Datenbanken sowie Google Scholar und Semantic Scholar. Eingeschlossen wurden Publikationen ab 2017 in deutscher oder englischer Sprache, die pflege- oder unterstützungsbedürftige Menschen und digitale Technologien zur Vernetzung innerhalb Sorgender Gemeinschaften thematisierten. Artikel, von denen nur Abstracts vorlagen, wurden ausgeschlossen. Die Recherche erfolgte anhand der PCC-Elemente: (1) Pflegepersonen, (2) Pflege- oder unterstützungsbedürftige Menschen, (3) Sorgende Gemeinschaften, (4) Digitale Technologien. Ergebnisse: Insgesamt wurden n = 105 Publikationen geprüft. Davon konnten n = 20 Publikationen in die Scoping-Überprüfung einbezogen werden. 15 Publikationen behandelten Modelle und Rahmenbedingungen von Sorgenden Gemeinschaften. 5 Artikel thematisierten digitale Technologien und Anwendungen für die Kommunikation und Vernetzung. Keine der Publikationen liefert Informationen über die Besonderheiten von Menschen mit Behinderungen. Schlussfolgerung: Es bestehen vielfältige Modelle und Rahmenbedingungen Sorgender Gemeinschaften. Die konzeptionelle Vielfalt und heterogenen Strukturen erschweren einen Vergleich der Modelle und ihrer Voraussetzungen. Partizipation, Governance und gemeinsame Verantwortung werden als zentrale Erfolgsfaktoren für die Implementierung beschrieben. Insgesamt lässt sich im deutschsprachigen Raum ein Trend hin zu partizipativen und gemeinwesenorientierten Versorgungsansätzen für unterstützungsbedürftige Menschen feststellen, die formelle und informelle Sorgeakteure miteinander vernetzen, wobei digitale Technologien die Kommunikation und Zusammenarbeit wesentlich unterstützen können.Introduction: The concept of “caring communities” describes an approach in which responsibility for elderly or care-dependent individuals is shared not only by professional caregivers but also by families, neighbors, volunteers, local organizations, and municipal authorities. Digital technologies can facilitate networking and communication within these communities, enabling collaboration across neighborhood boundaries. Methods: To explore and map the scope of the existing literature, a scoping review was conducted following the Joanna Briggs Institute (JBI) methodology for scoping reviews. A systematic search was carried out in four scientific databases, as well as in Google Scholar and Semantic Scholar. Inclusion criteria comprised publications in German or English from 2017 onward, addressing care-dependent individuals and digital technologies facilitating networking within caring communities. Publications limited to abstracts were excluded. The search strategy was structured using PCC elements: (1) caregivers, (2) individuals needing care, (3) caring communities, and (4) digital technologies. Results: A total of 105 publications were screened, of which 20 were included in the scoping review. Fifteen publications addressed models and contextual conditions of caring communities, while five articles focused specifically on digital technologies and applications in communication and networking. None of the publications provided specific information concerning the needs of individuals with disabilities. Conclusion: A variety of models of caring communities and their contextual conditions were identified. The conceptual diversity and heterogeneous structures make it difficult to compare the models and their prerequisites. Participation, governance, and shared responsibility are described as key success factors for implementation. Overall, there is an identifiable trend in German-speaking countries toward participatory, community-oriented care approaches for individuals with support needs, aiming to network formal and informal caregivers. Digital technologies have the potential to significantly enhance communication and collaboration within these networks.PeerReviewe
Soziale Arbeit und Polizei : Spannungen, Relationierungen und Interdependenze (Jahrgang 45, Heft 177)
No full textPeerReviewe
Wake turbulence reclassification and separation with induced power using the FAA aircraft characteristics database
Get PDFPurpose – The aim of this study is to reassess wake turbulence categorization (WTC) based on the calculation of induced power and to define separation minima grounded in established aircraft classifications during approach. This requires parameters such as aircraft mass, wingspan, approach speed, air density, and the Oswald factor. This approach is significantly more detailed than other metrics that consider only aircraft mass or a combination of mass and wingspan. --- Methodology – With the FAA Aircraft Characteristics Database, the necessary parameters for calculating induced power are accessible for 388 aircraft. This allows the definition of new thresholds for the HAW_WTC. Additionally, a first continuous separation formula was developed, derived from the structure of established separation charts by ICAO and Eurocontrol. --- Findings – Induced power can be derived also from the energy content of the vortex itself. Compared to ICAO and Eurocontrol schemes, the new wake turbulence classification offers a continuous, physics-based assessment of a specific aircraft's vortex generation potential. This continuous assessment enables aircraft-specific separation distances based on individual parameters for both the leader and follower aircraft. --- Research Limitations – Aircraft responses when flying directly through the vortex core or intersecting vortices at an angle are not considered in this study. --- Practical Implications – A physics-based wake turbulence categorization may offer more reliable assessments. Transitioning from generalized, fixed categories to a dynamic, real-time evaluation using live aircraft parameters could improve the efficiency and safety of flight operations. --- Originality – To date, no formula has been developed that allows for a continuous categorization and separation of aircraft during approach based on induced power. This work presents a first attempt to close that gap.NonPeerReviewe
Föderales Lernen am Beispiel der Kategorisierung von Mails
Get PDFZiel dieser Arbeit ist es, die praktische Umsetzung von föderalem Lernen zu untersuchen und die Vor- und Nachteile im Vergleich zum zentralisierten maschinellen Lernen zu betrachten. Dazu wird föderales Lernen mit dem Flower Framework anhand der Kategorisierung von Mails als Anwendungsfall umgesetzt. Zur Umsetzung der Kategorisierung werden Naive Bayes-Klassifikatoren, LSTM-Modelle und der Einsatz von Random Forest betrachtet. Zusätzlich zum föderalen Lernen werden Differential Privacy und Secure
Aggregation zum Schutz der Trainingsdaten eingesetzt. Das Training beim föderalen Lernen wird mit dem zentralen Lernen in verschiedenen Szenarien verglichen. Als Resultat der Arbeit wird erfolgreich aufgezeigt, wie die Umsetzung eines Anwendungsfalls beim föderalen Lernen erfolgen kann. Zudem kommt heraus, dass föderales Lernen allgemein Modelle mit ähnlicher Effektivität zu zentral trainierten Modellen trainieren kann, auch unter Einsatz von Differential Privacy und Secure Aggregation, und dass das Training durch die Verteilung des Trainings schneller abläuft.The goal of this thesis is to evaluate the practical implementation of federated learning and to examine the advantages and disadvantages in comparison to centralized machine learning. For this purpose, federated learning is implemented with the Flower Framework using the categorization of mails as a use case. Naive Bayes classifiers, LSTM models and the use of Random Forest are considered for the implementation of the categorization. In addition to federated learning, differential privacy and secure aggregation are used to protect the training data. The training with federated learning is compared with centralized learning in different scenarios. The result of the work is a successful demonstration of how a use case can be implemented in federated learning. It is also shown that federated learning can generally train models with similar effectiveness to centrally trained models, even when using differential privacy and secure aggregation, and that the training is faster due to the distribution of the training