Negotiated Content: Generative Soundscape Composition by Autonomous Musical Agents in Coming Together: Freesound

Generative music systems have been successful in styles and genres where there are explicit rules that can be programmed into the system. Practices and procedures within soundscape composition have tended to be implicit, in which recordings are selected, combined, and processed based upon contextual relationships. We present a system – Coming Together: Freesound – in which four autonomous artificial agents choose sounds from a large pre-analyzed database of soundscape recordings (from freesound.org), based upon their spectral content and metadata tags. Agents analyze, in realtime, other agentʼs audio, and attempt to avoid dominant spectral areas of other agents by selecting sounds that do not mask other agent’s spectra. Furthermore, selections from the database are constrained by metadata tags describing the sounds. Example compositions have been evaluated through subject testing, comparing them to human-composed compositions, and the results are discussed.