@article {GREEN:2024:0736-2935:3540, title = "A deep learning approach to predicting UAV sound affect", journal = "INTER-NOISE and NOISE-CON Congress and Conference Proceedings", parent_itemid = "infobike://ince/incecp", publishercode ="ince", year = "2024", volume = "270", number = "8", publication date ="2024-10-04T00:00:00", pages = "3540-3547", itemtype = "ARTICLE", issn = "0736-2935", url = "https://ince.publisher.ingentaconnect.com/content/ince/incecp/2024/00000270/00000008/art00061", doi = "doi:10.3397/IN_2024_3340", author = "GREEN, Marc and TORIJA MARTINEZ, Antonio", abstract = "The most prevalent method by which the impact of an acoustic environment upon people is assessed is the `environmental noise' approach, in which the sounds are quantified based entirely on their recorded weighted sound pressure levels. This does not account, however, for the character of sounds outside of their loudness. Objective metrics such as psychoacoustic annoyance have been developed to attempt to rectify this, however this is still heavily weighted towards loudness as the most important factor and may not provide a good estimate for subjective perceived annoyance for all sound sources, including unconventional aircraft such as unmanned aerial vehicles (UAVs). Here we present a novel method using a convolutional neural network (CNN) to predict perceived annoyance directly from mel-spectrograms derived from recorded audio featuring both traditional and unconventional aircraft sources. The network is trained and tested using data collated from five separate listening tests, and is shown to outperform conventional regression models based on psychoacoustic annoyance as a predictor of perceived annoyance. The time-frequency resolution of the calculated mel-spectrograms is shown to be a key factor in performance.", }