(4 min read)
Across Europe, cities are under growing pressure to balance quality of life with vibrant urban activity. The Slovenian coastal city of Koper, home to around 53,000 residents, exemplifies this challenge. With a dynamic mix of residential neighborhoods, tourist hotspots, and public events, Koper experiences diverse and fluctuating environmental noise levels. For municipalities seeking to protect urban well-being, understanding not only how much noise is generated, but also where it comes from and why, is essential.
The InnoRenew CoE demonstrator “Urban Noise Classification” introduces an innovative, privacy-preserving, and scalable IoT solution for monitoring and identifying urban noise. With the help of Swarmchestrate, a network of low-cost IoT devices is able to scale workload to achieve real time monitoring and classification.
A Smart City Challenge: Privacy, Performance, and Pressure
Traditional noise monitoring systems often follow a centralised architecture: microphones record sound, transmit the data to the cloud, and run machine learning algorithms to classify the source —be it traffic, construction, nightlife, or public events. While technically viable, this setup raises two major issues: first, it risks exposing raw audio to unauthorised access, challenging citizen privacy; second, the constant transmission of audio files consumes bandwidth and increases energy usage, making large-scale deployment costly and inefficient.
Swarmchestrate changes this equation.
Instead of relying on a centralised cloud infrastructure, the demonstrator leverages Swarmchestrate to orchestrate lightweight sound classification containers directly on edge devices. These devices —deployed across the city— process audio locally, ensuring that sensitive data never leaves the site where it is collected. This edge-based approach not only addresses privacy concerns, but also reduces communication overhead and improves response times.
From Noise to Insight —Without the Overhead
The system’s architecture is both robust and elegant. Each edge node is equipped with a Raspberry Pi, a noise meter, and an omnidirectional microphone. These nodes passively measure sound pressure levels throughout the day. When noise exceeds a defined threshold, a short audio clip is captured and queued for classification.
Custom-built, Swarmchestrate-branded, noise classification sensor (© Swarmchestrate consortium 2024-2026)
Crucially, not every node performs the classification task. Swarmchestrate continuously monitors the network’s health —including CPU load and thermal conditions on each device— and dynamically shifts classification workloads to available nodes. For example, if a device is operating in direct sunlight and nearing thermal throttling limits, its workload is rerouted to a cooler nearby node.
Once classification is complete, results (e.g., “traffic noise,” “crowd,” “music”) are sent —alongside raw noise levels— to a central server, where data is stored and visualised in a user-friendly dashboard accessible to city planners and stakeholders.
A Dashboard for Decision Makers
Koper’s city officials and urban planners are not expected to become data scientists overnight. The end-user interface developed as part of the demonstrator features a clean, interactive dashboard built with Grafana. It presents real-time plots of noise intensity across the city, overlays sound classification events on plots, and enables users to export data at various granularities for deeper analysis.
Urban noise classification dashboard for decision makers (© Swarmchestrate consortium 2024-2026)
Swarmchestrate offers a decentralised orchestration framework well-suited to the unpredictable and changing nature of urban environments. In Koper, its ability to balance computation workloads based on device constraints, while preserving citizen privacy, is a game-changer. As more cities across Europe turn to AI-driven monitoring systems, the InnoRenew’s demonstrator showcases how ethical, efficient, and sustainable smart city solutions can be designed from the ground up.
Beyond the City: A Blueprint for Responsible Sensing
Koper’s demonstrator is not just about urban noise. It reflects a broader vision: that smart sensing solutions must be adaptable, transparent, and above all, respectful of the communities they serve. The partnership between InnoRenew CoE and Swarmchestrate highlights how academic institutions and European innovation programs can work hand in hand with municipalities to pilot technology that is both cutting-edge and citizen-centric. However, there are many differences among the built environment in cities governing the quality of service (QoS) requirements. In Koper, the summer temperatures are quite high, coincidentally, summer is when the number of events is at its peak. In practical terms, this leads to excessive temperature buildup in the IP68 rated electrical box due to a high number of AI inferences. With the help of Swarmchestrate, inferences can migrate to other devices in the swarm that are hopefully not directly exposed to sunlight at the time in order to guarantee the QoS requirements.
The system is currently live in select zones of Koper, with plans to expand coverage in the coming months. Early feedback from city officials and residents has been overwhelmingly positive, validating the project’s emphasis on privacy, performance, and participatory design.
Editors: Aleksandar Tošić and Niki Hrovatin, InnoRenew CoE