Smooth Granular Sound Texture Synthesis by Control of Timbral Similarity
Résumé
Granular methods to synthesise environmental sound textures (e.g. rain, wind, fire, traffic, crowds) preserve the richness and nuances of actual recordings, but need a preselection of timbrally stable source excerpts to avoid unnaturally-sounding jumps in sound character. To overcome this limitation, we add a description of the timbral content of each sound grain to choose successive grains from similar regions of the timbre space. We define two different timbre similarity measures, one based on perceptual sound descriptors, and one based on MFCCs. A listening test compared these two distances to an uncon-strained random grain choice as baseline and showed that the descriptor-based distance was rated as most natural, the MFCC based distance generally as less natural, and the random selection always worst.
Origine : Fichiers produits par l'(les) auteur(s)
Loading...