Soundscape classifying ants

In this paper, the use of fuzzy ant clustering in classifying a large database of environmental soundscape recordings is outlined. Fuzzy ant clustering is a soft computing technique inspired by the clustering behaviour observed in colonies of several ant species. Virtual ants or “agents” move through the database, “pick up” soundscape recordings and drop them on places where similar recordings are present. Similarity of soundscape recordings is expressed by fuzzy resemblance of the shape of the SPL histogram, the frequency spectrum and the spectrum of temporal fluctuations, representing loudness, spectral and temporal content. The fuzzy IF-THEN rules, governing the behaviour of the virtual ants, are optimized using a specially adapted genetic algorithm, in order to achieve an optimal set of homogeneous clusters. Advantages of this approach, as compared to traditional clustering methods, are that no a priori information, such as the desired number of clusters, is needed, and that a more flexible set of indicators can be used. The clustering model is validated on a database of acoustic measurements of 1116 soundscapes, made in 16 urban parks in Stockholm, and results are compared with visitor survey data on soundscape quality.