Nearest-neighbor Generic Sound Classification with a WordNet-based Taxonomy

Audio classification methods work well when fine-tuned to reduced domains, such as musical instrument classification or simplified sound effects taxonomies. Classification methods cannot currently offer the detail needed in general sound recognition. A real-world-sound recognition tool would require a taxonomy that represents the real world and thousands of classifiers, each specialized in distinguishing little details. To tackle the taxonomy definition problem we use WordNet, a semantic network that organizes real world knowledge. In order to overcome the second problem, that is the need of a huge number of classifiers to distinguish a huge number of sound classes, we use a nearest-neighbor classifier with a database of isolated sounds unambiguously linked to WordNet concepts.