Semantic modelling for user interaction with sonic content

Abstract

In this paper we present a methodology for converting semantic descriptions of sounds into computable audio features. This process aims to enable the use of commonly used notions of timbre in an audio engineering context where the user interacts (e.g. searches for sounds in large digital collections) with sonic content, bridging the gap between the high-level perceptual sound notions and low-level machine-ready descriptors. Although our focus is the description of the constituent blocks of a generalpurpose semantic framework, examples from an experimental test for the semantic characterization of drum samples will be given to illustrate the process.