Lionello, Matteo; (2021) A new methodology for modelling urban soundscapes: a psychometric revisitation of the current standard and a Bayesian approach for individual response prediction. Masters thesis (M.Phil), UCL (University College London).

Preview

Text LIONELLO_MPHIL_FINAL.pdf - Accepted Version Download (41MB) | Preview

Abstract

Measuring how the urban sound environment is perceived by public space users, which is usually referred as urban soundscape, is a research field of particular in terest for a broad and multidisciplinary scientific community besides private and public agencies. The need for a tool to quantify soundscapes would provide much support to urban planning and design, so to public healthcare. Soundscape liter ature still does not show a unique strategy for addressing this topic. Soundscape definition, data collection, and analysis tools have been recently standardised and published in three respective ISO (International Organisation for Standardization) items. In particular, the third item of the ISO series defines the calculation of the soundscape experience of public space users by means of multiple Likert scales. In this thesis, with regards to the third item of the soundscape ISO series, the soundscape data analysis standard method is questioned and a correction paradigm is proposed. This thesis questiones the assumption of a point-wise superimposition match across the Likert scales used during the soundscape assessment task. In order to do that, the thesis presents a new method which introduces correction values, or metric, for adjusting the scales in accordance to the results of common scaling behaviours found across the investigated locations. In order to validate the results, the outcome of the new metric is used as tar get to predict the individual experience of soundscapes from the participants. In comparison to the current ISO output, the new correction values reveal to achievea better predictability in both linear and non-linear modelling by increasing the ac-curacy of prediction of individual responses up to 52.6% (8.3% higher than theaccuracy obtained with the standard method).Finally, the new metric is used to validate the collection of data samples acrossseveral locations on individual questionnaires responses. Models are trained, in aiterative way, on all the locations except the one used during the validation. Thisprocedure provides a strong validating framework for predicting individual subjectassessments belonging to locations totally unseen during the model training. The results show that the combination of the new metrics with the proposed modelling structure achieves good performance on individual responses across the dataset withan average accuracy above 54%. A new index for measuring the soundscape is fi-nally introduced based on the percentage of people agreeing on soundscape pleas-antness calculated from the new proposed metric and performing a r-squared valueequals to 0.87.The framework introduced is limited by cultural and linguistic factors. Indeed,different corrected metric space are expected to be found when data is collected from different countries or urban context. The current values found in this thesis areso expected to be valid in large British cities and eventually in international hub andcapital cities. In these scenarios the corrected metric would provide a more realisticand direction-invariant representation of how the urban soundscape is perceived compared to the current ISO tool, showing that some components in the circumplex model are perceived softer or stronger according to the dimension. Future research will need to understand better the limitations of this new ramework and to extendand compare it towards different urban, cultural, and linguistic contexts.

Download activity - last month

Export as JSONCSVXML

Download activity - last 12 months

Export as CSVXMLJSON

Downloads by country - last 12 months

Export as JSONXMLCSV

Archive Staff Only

View Item