A systematic review of prediction models for the experience of urban soundscapes
Matteo Lionello,Francesco Aletta,Jian Kang
DOI: https://doi.org/10.1016/j.apacoust.2020.107479
IF: 3.614
2020-01-01
Applied Acoustics
Abstract:A systematic review for soundscape modelling methods is presented. The methods for developing soundscape models are hereby questioned by investigating the following aspects: data acquisition methods, indicators used as predictors of descriptors in the models, descriptors targeted as output of the models, linear rather than non-linear model fitting, and overall performances. The inclusion criteria for the reviewed studies were: models dealing with soundscape dimensions aligned with the definitions provided in the ISO 12913 series; models based on soundscape data sampled at least at two different locations and using at least two variables as indicators. The Scopus database was queried. Biases on papers selection were considered and those related to the methods are discussed in the current study. Out of 256 results from Scopus, 22 studies were selected. Two studies were included from the references among the results. The data extraction from the 24 studies includes: data collection methods, input and output for the models, and model performance. Three main data collection methods were found. Several studies focus on the different combination of indicators among physical measurements, perceptual evaluations, temporal dynamics, demographic and psychological information, context information and visual amenity. The descriptors considered across the studies include: acoustic comfort, valence, arousal, calmness, chaoticness, sound quality, tranquillity, and vibrancy. The interpretation of the results is limited by the large variety of methods, and the large number of parameters in spite of a limited amount of studies obtained from the query. However, perceptual indicators, visual and contextual indicators, as well as time dynamic embedding, overall provide a better prediction of soundscape. Finally, although the compared performance between linear and non-linear methods does not show remarkable differences, non-linear methods might still represent a more suitable choice in models where complex structures of indicators are used.