Koper is a coastal city in Slovenia that balances residential life with tourism, traffic, and frequent public events. These pressures elevate ambient noise and call for continuous monitoring and real-time insight into noise sources to support data-informed urban planning. The current noise monitoring system operates on Raspberry Pi sensor nodes equipped with a sound-pressure-level (SPL) sensor and an omnidirectional microphone. When the SPL exceeds 70 dB, the node records a brief audio clip, converts it to a spectrogram via a Fast Fourier Transform (FFT), and sends this spectrogram to the cloud for noise-source classification. Classification currently runs in the cloud because state-of-the-art audio models are computationally demanding and sustained on-device inference can overheat Raspberry Pi hardware, degrading performance and reducing device lifetime. This cloud-centric strategy, however, carries drawbacks: residual privacy concerns and the inefficiency of transmitting substantial feature data to the cloud.

Added value by Swarmchestrate: Swarmchestrate upgrades the system by enabling true noise-source classification at the edge. It orchestrates containerised classifiers across the Raspberry Pi fleet, scaling them to maintain the required compute capacity and dynamically migrating them within the edge network to avoid device overheating. In this design, audio never leaves the edge network; only final classification results and essential metrics are sent to the backend. The outcome is stronger privacy, lower bandwidth and cloud costs, and faster response time. At the same time, the demonstrator showcases Swarmchestrate’s capability to coordinate containerised microservices seamlessly across the cloud-to-edge continuum, delivering accurate, near-real-time noise insights that help the Municipality of Koper manage noise more effectively.