Modelling sound preference

Abstract

More and more attention is being paid to soundscape quality by today’s acoustic researchers and environmental designers. It is important that, in addition to noise annoyance, sound preference is another main component to determine soundscape quality, especially in urban open public spaces. Previous studies showed that subject evaluation of sound preference in urban open public spaces varied according to physical, psychological, and social variables [W. Yang and J. Kang, J. Urban Des. 10, 69–88 (2005)]. Prediction models using artificial neural network (ANN) techniques for sound preference have been developed, as presented in this paper. Suitable input variables for ANN models have been derived from SPSS analyses based on data collected from large scale field studies and laboratory experiments [L. Yu and J. Kang, J. Acoust. Soc. Am. 120, 3238 (2006)]. These input variables include psychoacoustic indices such as loudness, sharpness and roughness, subject’s social/demographic factors and their long‐term sound experience, while the sound preference, described as acoustic comfort or pleasantness, is set as the output. Through the process of training and testing the models, good convergences have been achieved, which means that the prediction of sound preference evaluation is possible at the design stage.