Toward an evidence-based taxonomy of everyday sounds

An organizing account of everyday sounds could greatly simplify the management of audio data. The job of an audio database manager will typically involve assigning a combination of textual descriptors, and perhaps allocating to a predefined category. Retrieval is likely achieved by matching the descriptor to keyword search terms, or by browsing through categories. Whilst classification of musical instruments using this type of approach is relatively simple by virtue of the fact that a predefined taxonomy can follow a signal-related hierarchy, non-musical sounds do not necessarily follow such a hierarchy. In addition, classification is made more problematic by the ambiguity of words used to describe everyday sounds. Another area in which the issue of establishing a taxonomy of everyday sounds is particularly pertinent is that of soundscape research; research into soundscapes—acoustic environments as they are perceived—is a relatively new and multidisciplinary area, and as such descriptive terms for everyday sounds are currently used inconsistently. Many existing attempts to taxonomize everyday sounds prescribe categories that are not mutually exclusive, in that everyday sounds could exist in a number of categories; moreover, many are not based on empirical evidence. Here we present a robust method for creating an evidence-based taxonomy of everyday sounds, involving hierarchical clustering from dimensions generated by correspondence analysis of data from a simple card-sorting and naming procedur