Signal descriptors and temporal modeling for automatic sound classification

CUIDADO [16] and ECRINS [7] are new projects which aims at providing content-based music applications . Among these applications is an authoring tool for managing sample databases. It includes search by similarity, search by textual attributes, but also automatic sound classification based on predefined taxonomies and user defined taxonomies. This last point raises a crucial issue concerning the design of the classifier and the choice of the appropriate signal descriptors well suited for classification. This paper concentrates on the design of CUIDADO and ECRINS classifiers and on two algorithms for automatic selection of the most appropriate signal descriptors for a given taxonomy: based on discriminant analysis and on mutual information.