ImprovGenerator: Online Grammatical Induction for On-the-Fly Improvisation Accompaniment

Abstract

We propose an online generative algorithm to enhance musical expression via intelligent improvisation accompaniment. Our framework called the ImprovGenerator, takes a live stream of percussion patterns and generates an improvised accompaniment track in real-time to stimulate new expressions in the improvisation. We use a mixture model to generate an accompaniment pattern, that takes into account both the hierarchical temporal structure of the live input patterns and the current musical context of the performance. The hierarchical structure is represented as a stochastic context-free grammar, which is used to generate accompaniment patterns based on the history of temporal patterns. We use a transition probability model to augment the grammar generated pattern to take into account the current context of the performance. In our experiments we show how basic beat patterns performed by a percussionist on a cajon can be used to automatically generate on-the-fly improvisation accompaniment for live performance.