Predicting the Perceptual Rating of a Soundscape Using Artificial Intelligence

Abstract

The Sons al Balct'o project was a citizen science initiative that started during the lockdown consisting in collecting videos recorded in domestic balconies during 2020 and 2021, along with perceptual questionnaires about the soundscape. It has developed into a citizen science project that aims at analyzing the quality of urban quiet places. In previous work, we applied artificial intelligence techniques to automatically detect and classify acoustic events present on the videos. However, to better understand the impact of these events on people's lives, it is important to consider their perception of the soundscape. In this work, we propose a method to incorporate perceptual data into the acoustic event detection algorithm. Specifically, we aim to predict users' perception of the soundscape based on their responses to the questionnaires. Our target is to train a machine learning model to predict the perceptual rating of a soundscape based on the acoustic events and their acoustic features present in a location. When completed, our work will enable a more comprehensive and accurate assessment of the soundscape, taking into account both objective acoustic features and subjective perceptual responses.