Mitchell, Andrew James; (2022) Predictive Modelling of Complex Urban Soundscapes: Enabling an engineering approach to soundscape design. Doctoral thesis (Ph.D), UCL (University College London).

Preview

Text Andrew Mitchell PhD Thesis Final.pdf - Accepted Version Download (13MB) | Preview

Abstract

Conventional noise control methods typically limit their focus to the reduction of unwanted noise, ignoring the benets of positive sounds and struggling to reflect the totality of noise impacts. Modern approaches to achieve improved health outcomes and public satisfaction aim to incorporate the perception of an acoustic environment, an approach known as ‘soundscape’. When attempting to apply soundscape in practice, it is apparent that new methods of analysing soundscape perception in urban spaces are required; in particular, a predictive model of the users’ perceptual response to the acoustic environment is necessary. This thesis is intended to enable a move towards applying engineering approaches to soundscape design. This is achieved by developing predictive models of soundscape perception through empirical studies examining a large scale soundscape assessment database. The results are presented in three parts: first, the data collection protocol and modelling methods developed for this work are presented; the second part demonstrates an initial development and application of a predictive soundscape model; the final section expands upon this initial model with two empirical studies exploring the potential for additional information to be included in the model. This thesis begins by establishing a protocol for large scale soundscape data collection based on ISO 12913-2 and the creation of a database containing 1,318 responses paired with 693 binaural recordings collected in 13 locations in London and Venice. The first study then presents an initial development and application of a model designed to predict soundscape perception based on psychoacoustic analysis of the binaural recordings. Through the collection of an additional 571 binaural recordings during the COVID-19 lockdowns, sound level reductions at every location are seen, ranging from a reduction of 1.27 dB(A) in Regents Park Japan to 17.33 dB(A) in Piazza SanMarco, with an average reduction across all locations of 7.27 dB(A). Multi-level models were developed to predict the overall soundscape pleasantness (R2 = 0.85) and eventfulness (R2 = 0.715) of each location and applied to the lockdown recordings to determine how the soundscape perception likely changed. The results demonstrated that perception shifted toward less eventful soundscapes and to more pleasant soundscapes for previously traffic-dominated locations but not for human- and natural-dominated locations. The modelling process also demonstrated that contextual information was important for predicting pleasantness but not for predicting eventfulness. The next stage of the thesis considers a series of expansions to the initial model. The second piece of empirical work makes use of a dataset of recordings collected from a Wireless Acoustic Sensor Network (WASN) which includes sound source labels and annoyance ratings collected from 100 participants in an online listening study. A multilevel model was constructed using a combination of psychoacoustic metrics and sound source labels to predict perceived annoyance, achieving an R2 of 0.64 for predicting individual responses. The sound source information is demonstrated to be a crucial factor, as the relationship between roughness, impulsiveness, and tonality and the predicted annoyance varies as a function of the sound source label. The third piece of empirical work uses multilevel models to examine the extent to which personal factors influence soundscape perception. The findings suggest that personal factors, including psychological wellbeing, age, gender, and occupational status, account for approximately 1.4% of the variance for pleasantness and 3.9% for eventfulness, while the influence of the locations accounted for approximately 34% and 14%, respectively. Drawing from the experience gained working with urban soundscape data, a new method of analysing and presenting the soundscape perception of urban spaces is developed. This method inherently considers the variety of perceptions within a group and provides an open-source visualisation tool to facilitate a nuanced approach to soundscape assessment and design. Based on this empirical evidence, a framework is established for developing future predictive soundscape models which can be integrated into an engineering approach. At each stage, the results of these studies is discussed in terms of how it can contribute to a generalisable predictive soundscape model.

Download activity - last month

Export as CSVXMLJSON

Download activity - last 12 months

Export as JSONXMLCSV

Downloads by country - last 12 months

Export as CSVXMLJSON

Archive Staff Only

View Item