Acoustic comfort prediction in urban environments based on automatic sound events detection and noise levels obtained from a wireless acoustic sensors network

There is a growing concern about the high levels of noise pollution in many urban environments across the world and, more specifically, in metropolises such as Barcelona. Prolonged noise exposure has adverse health and wellbeing effects. Therefore, there is a growing interest in enforcing measures to reduce this negative impact in the citizen’s quality of life. The first issue that must be tackled is finding a method to accurately evaluate the perceived acoustic comfort or discomfort in a given location, as there is often a wide variety of factors affecting the perception of a soundscape.

 There are currently different approaches to predict the acoustic comfort level using several indicators, i.e., acoustic, psycho-acoustic, non-auditory and even non-sensory. The main limitation of these proposals is that they are grounded in costly (even inviable in some contexts) data collection processes. In many cases, the methods proposed require using specialised and calibrated equipment (eg. sonometers or spectrum analyzers) on-site. This kind of equipment is expensive and it needs some level of expertise to use it properly. In some other cases, they are based on a collective assessment conducted by several participants who provide their opinion on a given soundscape. Some studies go as far as to recommend the inclusion of specific information about the listener who will be exposed to the evaluated soundscape (information that not always is available). 

 The main goal of this research is to propose a new method to predict the acoustic comfort around a given urban location using the few amount of data that a single person can collect without neither specialised equipment nor technical background. First, an automatic sound events classifier has been designed and set to work with polyphonic real data obtained from a short recording acquired through a smartphone. Subsequently, an analysis has been performed on the viability of adding data from a wireless acoustic sensors network as indicators. Specifically, data from two distinct contexts obtained from Barcelona and Girona has been used: (1) data obtained during the lockdown caused by the COVID-19 pandemic and (2) data from a normal, free from mobility and activity restrictions, situation. Possible correlations between sensors’ data and the subjective perception of soundscapes located some distance apart have been explored. Finally, a predictor of the acoustic comfort has been implemented, based on the type of sounds automatically detected from the recording, complemented with the noise levels provided by the nearest sensor (from the wireless acoustic sensors network in the city) when they are available. Accuracies over 80% have been achieved for the acoustic comfort prediction using only a short audio clip recorded with a smartphone and the corresponding geo-localization coordinates.

 Overall, the main contributions of this research are: (1) the design of an automatic sound events classifier working on polyphonic audios obtained through smartphones, (2) the design of an acoustic comfort predictor based only on a short audio recording and data provided through a wireless acoustic sensors network, and (3) the exhaustive analysis of the changes caused in Barcelona’s soundscape during the COVID-19 lockdown.