We review a proposal made in the framework of MPEG-7 for the description of instrument sounds based on perceptual features. Results derived from experiments dealing with the perception of musical timbre allow us to approximate the main underlying perceptual dimensions from signal descriptors. A combination of these signal descriptors allows a comparison between sounds in terms of an estimated perceived timbral (dis)similarity. Since the proposed model is based on experiments using mainly synthesized sounds, a validation of the model was performed on a database composed of natural musical sounds.
Contribution au colloque ou congrès : ICMC: International Computer Music Conference