Comparison of Deep Learning and Psychoacoustic Models to Predict UAV Noise Impact in Soundscapes

Ellis, Max W; Green, Marc C; Lotinga, Michael J B; Torija Martínez, Antonio J

Abstract

* The increasing prevalence of Unmanned Aircraft Systems in urban environments necessitates a deeper understanding of their impact on the experience of urban soundscapes. This study presents Machine Learning models aimed at predicting perceived annoyance of UAS noise. Deep learning models were generated using convolutional recurrent neural networks, trained on a dataset incorporating data from multiple listening experiment. The model predictions are compared with various existing nonlinear models for Psychoacoustic Annoyance. Our expanded dataset includes recent field studies across England and Greece, enhancing the robustness and generalisability of our models. The broader aim of this research is development of a comprehensive soundscape model for UAS noise, which could be incorporated into future 'next generation' smart sound level meters and be used to inform urban planning decisions.

• Presentation Conference Type: Conference Paper (published)
• Conference Name: Forum Acusticum EuroNoise 2025
• Start Date: Jun 23, 2025
• End Date: Jun 26, 2025
• Acceptance Date: Apr 27, 2025
• Publication Date: 2025-06
• Deposit Date: Jun 21, 2025
• Journal: 11 th Convention of the European Acoustics Association
• Peer Reviewed: Peer Reviewed
• Keywords: Artificial Intelligence; Machine Listening; Psychoacoustic Annoyance; Soundscape; UAS