Evolutionary Music: Statistical Learning and Novelty for Automatic Improvisation

In this work we combine aspects of implicit learning with novelty search in an evolutionary algorithm with the aim to automatically generate melodies with improvisational flavour. Using Markov chains, the technique we present combines implicit statistical knowledge, extracted from musical corpora, with an adaptive novelty search mechanism. The algorithm is described along with the main design choices. Preliminary results are shown in two different musical contexts: Irish music and counterpoint compositions