TY - JOUR
T1 - Prediction of the acoustic comfort of a dwelling based on automatic sound event detection
AU - Bonet-Solà, Daniel
AU - Vidaña-Vila, Ester
AU - Alsina-Pagès, Rosa Ma
N1 - Publisher Copyright:
© 2023 the author(s).
PY - 2023/1/1
Y1 - 2023/1/1
N2 - There is an increasing concern about noise pollution around the world. As a first step to tackling the problem of deteriorated urban soundscapes, this article aims to develop a tool that automatically evaluates the soundscape quality of dwellings based on the acoustic events obtained from short videos recorded on-site. A sound event classifier based on a convolutional neural network has been used to detect the sounds present in those videos. Once the events are detected, our distinctive approach proceeds in two steps. First, the detected acoustic events are employed as inputs in a binary assessment system, utilizing logistic regression to predict whether the user’s perception of the soundscape (and, therefore, the soundscape quality estimator) is categorized as “comfortable” or “uncomfortable”. Additionally, an Acoustic Comfort Index (ACI) on a scale of 1–5 is estimated, facilitated by a linear regression model. The system achieves an accuracy value over 80% in predicting the subjective opinion of citizens based only on the automatic sound event detected on their balconies. The ultimate goal is to be able to predict an ACI on new locations using solely a 30-s video as an input. The potential of the tool might offer data-driven insights to map the annoyance or the pleasantness of the acoustic environment for people, and gives the possibility to support the administration to mitigate noise pollution and enhance urban living conditions, contributing to improved well-being and community engagement.
AB - There is an increasing concern about noise pollution around the world. As a first step to tackling the problem of deteriorated urban soundscapes, this article aims to develop a tool that automatically evaluates the soundscape quality of dwellings based on the acoustic events obtained from short videos recorded on-site. A sound event classifier based on a convolutional neural network has been used to detect the sounds present in those videos. Once the events are detected, our distinctive approach proceeds in two steps. First, the detected acoustic events are employed as inputs in a binary assessment system, utilizing logistic regression to predict whether the user’s perception of the soundscape (and, therefore, the soundscape quality estimator) is categorized as “comfortable” or “uncomfortable”. Additionally, an Acoustic Comfort Index (ACI) on a scale of 1–5 is estimated, facilitated by a linear regression model. The system achieves an accuracy value over 80% in predicting the subjective opinion of citizens based only on the automatic sound event detected on their balconies. The ultimate goal is to be able to predict an ACI on new locations using solely a 30-s video as an input. The potential of the tool might offer data-driven insights to map the annoyance or the pleasantness of the acoustic environment for people, and gives the possibility to support the administration to mitigate noise pollution and enhance urban living conditions, contributing to improved well-being and community engagement.
KW - acoustic comfort
KW - acoustic event detection
KW - annoyance evaluation
KW - citizen science
KW - convolutional neural networks
KW - noise
KW - soundscape
UR - http://www.scopus.com/inward/record.url?scp=85182153902&partnerID=8YFLogxK
U2 - 10.1515/noise-2022-0177
DO - 10.1515/noise-2022-0177
M3 - Article
AN - SCOPUS:85182153902
SN - 2084-879X
VL - 10
JO - Noise Mapping
JF - Noise Mapping
IS - 1
M1 - 20220177
ER -