Automatic Construction of Interactive Machine Improvisation Scenarios from Audio Recordings

We describe a system that allows improvisers and composers to construct an interactive musical environment directly from a musical recording. Currently, interactive music pieces require separate phases of constructing generative models and structuring them into a larger compositional plan. In the proposed system we combine machine improvisation tools based on Variable Markov Oracle (VMO) with an interactive score (i-score) to control the improvisation according to larger structures found in that recording. This allows construction of improvisation scenarios in ways that are organic with the musical materials used for generating the music. The method uses new results of audio segmentation based on VMO and translates it into a Petri Net (PN) model with transition rules left open to be defined by a musician. The PN structure is finally translated into a timed representation for a live i-score control.