A deformable musical instrument can take numerous dis- tinct shapes with its non-rigid features. Building audio syn- thesis module for such an interface behaviour can be chal- lenging. In this paper, we present the Al-terity, a non-rigid musical instrument that comprises a deep learning model with generative adversarial network architecture and use it for generating audio samples for real-time audio synthesis. The particular deep learning model we use for this instru- ment was trained with existing data set as input for pur- poses of further experimentation. The main benefits of the model used are the ability to produce the realistic range of timbre of the trained data set and the ability to generate new audio samples in real-time, in the moment of playing, with the characteristics of sounds that the performer ever heard before. We argue that these advanced intelligence features on the audio synthesis level could allow us to ex- plore performing music with particular response features that define the instrument’s digital idiomaticity and allow us reinvent the instrument in the act of music performance.

Peer reviewed