Sonic Xplorer: A Machine Learning Approach for Parametric Exploration of Sound

This paper presents Sonic Xplorer an interfaces that uses timbre adjectives for multiparametric control sound synthesis. The interface utilises an artificial neural network to create a personalised interface. Users can manipulate a large number of sound synthesis parameters without the need to learn or use the synthesiser’s complex interface by utilising programed sounds by expert users. Sonic Xplorer learns a correlation based on users’ ratings between timbre adjectives and the acoustic descriptors. Timbre adjectives are then used to describe the acoustic qualities of the desired sound. This paper discusses in detail the approach that has been followed to develop the system and the mapping and strategies users employed when using the interface in order to discover new sounds.