Artificial neural network models of sound signals in urban open spaces

Abstract

Sound signals, known as foreground sounds, are important components of soundscape in urban open spaces. Previous studies in this area have shown that sound preferences are different according to social and demographic factors of individual users [W. Yang and J. Kang, J. Urban Des., 10, 69–88 (2005)]. This study develops artificial neural network (ANN) models of sound signals for architects at design stage, simulating subjective evaluation of sound signals. A database for ANN modeling has been established based on large-scale social surveys in European and Chinese cities. The ANN models have consequently been built, where individual’s social and demographic factors, activities, and acoustic features of the space and sounds are used as input variables while the sound preference is defined as the output. Through the process of training and testing the ANN models, considerable convergences have been achieved, which means that the models can be applied as practical tools for architects to design sound signals in urban open spaces, taking the characteristics of potential users into account. Currently ANN models combining foreground and background sounds are being developed.