Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Real-time Prediction of Urban Sound Propagation with Conditioned Normalizing Flows

Achim Eckerle, Martin Spitznagel, Janis Keuper

TL;DR

The paper tackles the need for real-time, physically consistent urban sound propagation for regulatory and planning workflows. It introduces a fully conditioned normalizing-flow model (Full-Glow) that maps 2D city layouts to CNOSSOS-compliant sound-pressure maps with high fidelity across Baseline, Diffraction, and Reflection regimes, while delivering over a 2000× speedup on a single RTX 4090. The approach achieves state-of-the-art NLoS accuracy (e.g., Baseline MAE ≈ 0.65 dB) and preserves complex wave phenomena such as diffraction and interference, enabling interactive exploration for urban planning and compliance mapping. The work demonstrates practical impact by providing a fast, stable, and scalable engine suitable for scenarios like road closures and night-work assessments, with robust statistical significance across diverse acoustic conditions.

Abstract

Accurate and fast urban noise prediction is pivotal for public health and for regulatory workflows in cities, where the Environmental Noise Directive mandates regular strategic noise maps and action plans, often needed in permission workflows, right-of-way allocation, and construction scheduling. Physics-based solvers are too slow for such time-critical, iterative "what-if" studies. We evaluate conditional Normalizing Flows (Full-Glow) for generating for generating standards-compliant urban sound-pressure maps from 2D urban layouts in real time per 256x256 map on a single RTX 4090), enabling interactive exploration directly on commodity hardware. On datasets covering Baseline, Diffraction, and Reflection regimes, our model accelerates map generation by >2000 times over a reference solver while improving NLoS accuracy by up to 24% versus prior deep models; in Baseline NLoS we reach 0.65 dB MAE with high structural fidelity. The model reproduces diffraction and interference patterns and supports instant recomputation under source or geometry changes, making it a practical engine for urban planning, compliance mapping, and operations (e.g., temporary road closures, night-work variance assessments).

Real-time Prediction of Urban Sound Propagation with Conditioned Normalizing Flows

TL;DR

The paper tackles the need for real-time, physically consistent urban sound propagation for regulatory and planning workflows. It introduces a fully conditioned normalizing-flow model (Full-Glow) that maps 2D city layouts to CNOSSOS-compliant sound-pressure maps with high fidelity across Baseline, Diffraction, and Reflection regimes, while delivering over a 2000× speedup on a single RTX 4090. The approach achieves state-of-the-art NLoS accuracy (e.g., Baseline MAE ≈ 0.65 dB) and preserves complex wave phenomena such as diffraction and interference, enabling interactive exploration for urban planning and compliance mapping. The work demonstrates practical impact by providing a fast, stable, and scalable engine suitable for scenarios like road closures and night-work assessments, with robust statistical significance across diverse acoustic conditions.

Abstract

Accurate and fast urban noise prediction is pivotal for public health and for regulatory workflows in cities, where the Environmental Noise Directive mandates regular strategic noise maps and action plans, often needed in permission workflows, right-of-way allocation, and construction scheduling. Physics-based solvers are too slow for such time-critical, iterative "what-if" studies. We evaluate conditional Normalizing Flows (Full-Glow) for generating for generating standards-compliant urban sound-pressure maps from 2D urban layouts in real time per 256x256 map on a single RTX 4090), enabling interactive exploration directly on commodity hardware. On datasets covering Baseline, Diffraction, and Reflection regimes, our model accelerates map generation by >2000 times over a reference solver while improving NLoS accuracy by up to 24% versus prior deep models; in Baseline NLoS we reach 0.65 dB MAE with high structural fidelity. The model reproduces diffraction and interference patterns and supports instant recomputation under source or geometry changes, making it a practical engine for urban planning, compliance mapping, and operations (e.g., temporary road closures, night-work variance assessments).

Paper Structure

This paper contains 16 sections, 2 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Physics-Based Urban Acoustics.
  4. Deep Learning for Physics Simulation.
  5. Normalizing Flows.
  6. Method
  7. Data.
  8. Fully conditioned Glow.
  9. Likelihood objective.
  10. Training setup.
  11. Metrics.
  12. Results
  13. Quantitative Analysis.
  14. Qualitative and Structural Analysis.
  15. Statistical Significance.
  16. ...and 1 more sections

Figures (3)

  • Figure 1: Example data pair: (a) Input urban layout from OSM, and corresponding ground truth simulations for (b) Baseline, (c) Diffraction, and (d) Reflection scenarios.
  • Figure 2: Visual comparison of our model's predictions (center column) against the ground truth (left column) for the Baseline, Reflection, and Diffraction scenarios. The absolute error maps (right column) confirm high physical fidelity across all cases.
  • Figure 3: Model comparison with 95% confidence intervals across all three acoustic scenarios, computed over 1,245 test samples per scenario. The non-overlapping confidence intervals confirm the statistical significance of our Full-Glow model's performance gains, particularly in acoustically shadowed (NLoS) regions.