Urban noise management often suffers from a gap between broad, city-level policies and the street-level conditions actually experienced by residents. This thesis develops and validates an \emph{interpretable} machine-learning framework that combines street-view imagery (SVI) with on-site acoustic measurements to predict both a standard physical metric (A-weighted equivalent sound level, LAeq) and a psychoacoustic metric (Zwicker Loudness), which reflects how loud sounds are perceived by people. The approach integrates advanced computer-vision feature extraction with ensemble learning, and uses interpretable AI tools (e.g., SHAP) to show how specific visual characteristics of the streetscape—such as vegetation cover, road proportion, building facades, and scene perception scores—are linked to predicted noise outcomes.

Tested across several Dutch cities, the models produce consistent, street-level predictions, enabling high-resolution \emph{diagnostic noise maps} for both LAeq and Loudness. Building on these maps, the thesis introduces a policy translation framework aligned with the Dutch \emph{Omgevingswet} and the EU Environmental Noise Directive (END). This framework includes: (i) identifying noise “hotspots” using both absolute thresholds (from WHO guidelines) and relative, within-city rankings; (ii) diagnosing the main visual features driving noise levels, using local model explanations; (iii) selecting targeted interventions—such as traffic flow adjustments, façade and surface treatments, and nature-based solutions—supported by documented mechanisms and measurable indicators; and (iv) establishing an update loop for periodic review as new imagery and measurements become available. An illustrative micro-case demonstrates how this process turns model outputs into actionable planning decisions and performance metrics.

The study's contributions are threefold: (1) an end-to-end, interpretable pipeline linking SVI and acoustics; (2) a dual-metric evaluation (LAeq and Zwicker Loudness) that combines legal compliance with a perception-based perspective; and (3) a concrete, regulator-aligned pathway from predictions and explanations to policy action. Limitations include the sample size relative to feature dimensionality, the absence of direct GIS or morphological data integration, and the geographic focus on Dutch cities. These factors point to future work involving larger and longitudinal datasets, multi-sensor/GIS integration, and transfer learning for broader applicability, while encouraging cautious, phased adoption in real-world planning.

This thesis explores the development of lightweight neural network models for predicting perceptual soundscape attributes and environmental sound sources, with the goal of enabling real-time, sensor-based soundscape monitoring in urban environments. Recognizing the limitations of conventional noise metrics, the study adopts a perceptual approach that accounts for how individuals experience urban soundscapes. The work leverages the Affective Responses to Augmented Urban Soundscapes (ARAUS) dataset to replicate and extend the training pipeline of the baseline AD_CNN model. Four parameter-reduction strategies are implemented - shrinking convolutional and dense layers, reducing temporal resolution, and increasing pooling - to produce compact models suitable for low-cost, resource-constrained sensors. Two optimized variants, AD_CNN_dense_layer and AD_CNN_hop_length, are combined into an ensemble model with only ~160K parameters, which outperforms the original AD_CNN in perceptual attribute prediction. A generalisation study evaluates model robustness on a separate urban parks dataset, revealing that while the lightweight ensemble performs well overall, it struggles with certain attributes such as pleasantness. Larger models like SoundAQnet generalise better but remain imperfect. The findings demonstrate the technical feasibility of lightweight perceptual models but also highlight the need for localised fine-tuning to account for subjective and context-dependent factors shaping soundscape perception. This work contributes to scalable, context-aware soundscape monitoring solutions for healthier urban environments.