Turkish Air Force Academy

Advances in synthetic speech generation pose significant risks to voice communication security. This study proposes an empirical framework for synthetic speech detection with an emphasis on cross-language robustness. Acoustic and prosodic features, including MFCC, CQCC, pitch, jitter, and shimmer, are evaluated using classical machine learning methods and deep learning models. Experiments are conducted on the FoR (monolingual) and ODSS (multilingual) benchmark datasets using stratified 80%–20% train–test splits and stratified 5-fold cross-validation. On FoR, a CNN achieves 99.2% accuracy with a 0.75% Equal Error Rate (EER). On ODSS, ResNet reaches 96.5% accuracy and 3.72% EER under the percentage split. Additionally, Random Forest attains 100% accuracy and 0% EER in 5-fold cross-validation on ODSS, indicating strong feature separability while suggesting potential overfitting effects. The results demonstrate that the proposed hybrid features and deep architectures provide high generalizability across different languages and synthesis techniques.

Lojistik sektöründe verimlilik, zaman yönetimi ve kalite faktörleri günümüzde işletmelerin rekabet gücünü belirleyen temel unsurlar hâline gelmiştir. Bu kapsamda, yalın yönetim yaklaşımlarından biri olan 5S yöntemi; çalışma alanlarında düzen, temizlik ve standardizasyon sağlayarak israfı azaltmayı ve operasyonel mükemmelliği geliştirmeyi amaçlamaktadır. Bu araştırmanın amacı, 5S uygulamalarının hava kargo taşımacılığına yönelik bir lojistik depo ortamındaki etkinliğini incelemektir. Çalışma, İstanbul’da faaliyet gösteren bir hava yolu kargo şirketinin lojistik merkezinde gerçekleştirilmiş olup, nitel ve nicel veri toplama yöntemleri bir arada kullanılmıştır. Saha gözlemleri ve denetim kontrol formlarından elde edilen veriler, 5S sisteminin doğru ve sürdürülebilir biçimde uygulanmasının iş kazalarını azalttığını, zaman kayıplarını önlediğini, çalışan memnuniyetini artırdığını ve genel verimliliği yükselttiğini ortaya koymaktadır. Sonuç olarak, hava kargo lojistik merkezlerinde 5S sistemi yalnızca fiziksel düzenin sağlanmasında değil, aynı zamanda kurumsal kültürün güçlendirilmesinde de stratejik bir yönetim aracı olarak değerlendirilmektedir.

Building fires are considered major disasters because of their significant effects on people, property, and the environment. This understanding has led to increased attention on developing preventive measures, particularly through the creation of effective methods for assessing fire risk. However, the effectiveness of these methods relies heavily on detailed physical and functional information of the building and data-driven decision-making. Building Information Modeling (BIM) has proven effective in representing structures, even in three dimensions. When integrated with Geographic Information Systems (GIS), it enhances spatial intelligence, leading to improved decision-making through robust multi-criteria approaches. Hence, this study develops a framework to assess fire risk in an indoor environment that deploys a BIM-based GIS and Multi-Criteria Decision-Making; this is specifically known as Technique for Order Preference by Similarity to Ideal Solution (TOPSIS). The developed framework consists of four steps: identifying fire risk parameters, calculating weights, conducting spatial fire risk assessments, and visualizing the results, where the developed concepts are tested and validated. According to the significant findings, the developed framework estimates that 18% of building rooms are at moderate risk, while the compared model identifies only 1%. This considerable difference could potentially arise from the detailed data structure of BIM and the spatial insights gained from GIS. By implementing the designed framework, key fire risk factors can be identified in three dimensions, accompanied by a comprehensive quantitative evaluation platform for fire risks within indoor environments.

In this study, we introduce two machine and deep learning based approaches to network traffic classification using sub-flows and sliding windows for near-real-time use-cases such as providing QoS/QoE to application flows. We develop two approaches: (1) a packet-based sub-flow method that uses a fixed number of packets for each instance, and (2) a time-based subflow method that utilizes packets in a fixed time range for each instance. Unlike similar studies conducted in the literature, our methods are designed to be applicable to real-time use cases where the classification should give an accurate answer quickly after the start of a network traffic to provide the necessary QoS/QoE to the network flows. We have conducted experiments with both methods considering the efficient window sizes and sliding intervals. We present the results of two classification goals: (1) application identification and (2) application category identification. Our results show that the proposed time-based approach provides at least 89% accuracy in application identification (depending on the application), and 96% in application category identification which makes it highly accurate for use-cases to provide QoS/QoE to application flows.