NN Music: Improvising with a 'Living' Computer

This paper proposes attributes of a living computer music, the product of a live algorithm. It illustrates how these attributes can inform creative design with reference to a real-time system for solo performer-machine collaboration, Neural Network Music, and the PQƒ framework proposed for live algorithms. Improvisation is treated as a classification problem at a high level of musical behaviour which can be measured statistically and train a multilayer perceptron neural network. Network outputs shape a stochastic-based synthesis engine. Mappings are covertly assigned, revisited by both player and machine as a performance develops. As the timing and choice of mapping is unknown, both participants are invited to learn and adapt to a responsive sonic environment which is created afresh on each performance. This offers a novel real-time application of feed-forward neural networks and a challenging, creative technological platform for freely improvised music