An Artificial Intelligence Approach to Sound Design


Session Title:

  • Composition (The Next Generation)

Presentation Title:

  • An Artificial Intelligence Approach to Sound Design



  • Abstract

    We consider music composition not only as the putting together of pre-existing sounds, but also composing (i.e. synthesizing) the sounds themselves. When synthesizing sounds composers have an intuition about the depth of possibilities of how these sounds should be organized into a musical structure. In attempting to obtain the desired sound the composer explores a variety of possible solutions trying out possibilities within a certain personal aesthetic style or idiom.

    New computer technology offers the most detailed control of the internal parameters of  synthesized sounds. Nowadays composers have the opportunity to deal with an infinite variety of sounds with considerable precision. However the effective use of this new technology increases the complexity of the compositional task. The computer can theoretically be programmed to synthesize any imaginable sound. This can get composers into trouble though. Besides the ability to design the appropriate synthesis program, he or she also needs to master a programming language in order to implement it and to ‘play’ with it. In such a situation we come to believe that musical imagination becomes vulnerable to extra-musical computer technicalities. Our research work in Artificial Intelligence-aided sound design is working towards a system which explores the possibility of using the computer as an assistant which works co-operatively with the composer. A prototype system, called ARTIST (for ARTficial Intelligencebased Synthesis Tool), has been implemented so far. Here the composer is able to design sounds by means of sound descriptions using English rather than by means of low level computer programming and numerical coding. In order to offer AI-based support for sound design we provided the system with machine learning mechanisms which, on the one hand, allow the user to teach ARTIST the vocabulary for sound descriptions and, on the other hand, enable ARTIST to automatically update its knowledge about sounds through user interaction.