Towards a Deep Improviser: a prototype deep learning post-tonal free music generator

Two modest-sized symbolic corpora of post-tonal and post-metrical keyboard music have been constructed, one algorithmic, the other improvised. Deep learning models of each have been trained. The purpose was to obtain models with sufficient generalisation capacity that in response to separate fresh input seed material, they can generate outputs that are statistically distinctive, neither random nor recreative of the learned corpora or the seed material. This objective has been achieved, as judged by k-sample Anderson-Darling and Cramer tests. Music has been generated using the approach, and preliminary informal judgements place it roughly on a par with an example of composed music in a related form. Future work will aim to enhance the model such that it deserves to be fully evaluated in relation to expression, meaning and utility in real-time performance.