IoT-AED

UrbanSound8K was selected due to its widespread adoption and balanced taxonomy, enabling comparability with prior work. To better reflect deployment scenarios, classes were adapted and reorganized, and the number of samples per class was adjusted through controlled balancing and targeted augmentation.

The resulting dataset captures realistic variability in environmental noise, recording conditions, and event occurrence. Each sample is accompanied by structured metadata, enabling leakage-safe group-based partitioning for reproducible evaluation. This dataset directly supports the experimental research presented in “Deep Learning-Based Acoustic Event Detection in IoT Edge Devices,” enabling realistic, reproducible, and edge-aware model evaluation.