More Cocreative Interaction : Pulse-Based Improvisation with Djazz 2.0
Résumé
Djazz 2.0 is an artificial intelligence system for human-machine co-improvisation in the context of pulse-based music. It can learn from the musical stream played by a human instrumentalist and generate stylistically coherent streams, synchronized to a given beat and respecting the musical constraints of a pre-existing scenario (e.g., chord progression, rhythmic pattern). Since pulse-based music is closely linked to bodily practices (rhythmic hand clapping, dancing), its development involves extensive ethnomusicological research on the social context of these practices. Djazz 2.0 has very little a priori musical knowledge and can adapt without bias to any particular cultural context, but it relies heavily on its user's knowledge of these contexts. Therefore, Djazz 2.0 has been designed as a “smart” digital instrument exploring a wide range of precise controls in real time. This work is the result of the European Research Council's REACH (Raising Cocreativity in Cyber–Human Musicianship) project, introduced in a previous article in the Computer Music Journal. The present article describes Djazz 2.0 in detail, from its anthropological challenges to its system architecture in Max/MSP, its theoretical learning and generation model using the factor oracle, its Antescofo-based tempo following, and its interaction strategies.