Augmenting Parametric Synthesis with Learned Timbral Controllers
Jeff Gregorio, Youngmoo E. Kim · Zenodo (CERN European Organization for Nuclear Research) · 2019
Feature-based synthesis applies machine learning and signal processing methods to the development of alternative interfaces for controlling parametric synthesis algorithms. One approach, geared toward real-time control, uses low dimensional gestural controllers and learned mappings from control spaces to parameter spaces, making use of an intermediate latent timbre distribution, such that the control space affords a spatially-intuitive arrangement of sonic possibilities. Whereas many existing systems present alternatives to the traditional parametric interfaces, the proposed system explores ways in which feature-based synthesis can augment one-to-one parameter control, made possible by fully invertible mappings between control and parameter spaces.