Musical variation and improvisation based on multi-resolution representations

Abstract

Musical creativity is one of the show pieces of Artificial Intelligence and during the last decades, many paths have been explored to capture musical style and to generate new music. One of the approaches exploits the similarities between music and language. In this paper, Fluid Construction Grammar, a state-of-the-art computational grammar is used to parse/analyse an existing piece, in order to create a variation on the song and generate an improvisation in the same style, using the same bi-directional grammar. A novel multi-resolution time representation to model musical melodies is presented.