Development of psychoacoustic prediction models for short-term noise annoyance responses to unmanned aircraft systems

Unmanned aircraft systems (UAS) are emerging for use in civil applications such as commercial logistics, surveying, agriculture, and maintenance tasks. One challenge raised by this technology is to understand how humans respond to UAS sound, the characteristics of which can be varied and unfamiliar, compared with conventional vehicles. Efforts are under way to facilitate flight path planning and optimisation incorporating noise annoyance prediction models. Coupled with UAS sound emission and propagation models, perception and response could be predicted using acoustic and psychoacoustic metrics found to be associated with subjective evaluation. However, identifying the most effective metrics and models is complicated by the wide array of possible descriptors. In this study, a multi-stage modelling approach was developed. This combined a flexible, non-parametric machine learning technique to identify acoustic and psychoacoustic metrics associated with noise annoyance responses to UAS sound, obtained experimentally within immersive audio scenes. This information was used to develop semi-parametric models to predict responses while addressing cluster-correlation in the data. This technique is contrasted with an alternative multilevel, mixed-effects regression approach to highlight the potential advantages. The prediction models also demonstrate how sound intensity, ambient acoustic environments, sound qualities, and number of flights affect UAS noise annoyance.