Representation and manipulation of sound textures for electronic music
Résumé
Sound textures and noises are sounds for that the understanding of the perceptually relevant features has remained unclear for a rather long time. This has changed with the work of Josh McDermott who proposed and perceptually validated a hypothesis about the perceptually relevant features, and at the same time proposed a method allowing to convert textures and noises by means of manipulating these features.
The presentation will introduce the set of features and demonstrate a software implementation for sound manipulation working with the set of features proposed by McDermott. Further, the representation will discuss our recent result of using convolutional neural networks for improved sound textures and noise synthesis.