Artificial Intelligence and Real-Time Interactive Improvisation

My long-term goal is to interactively improvise with the computer, using it as a tool when I practice at home alone in order to help me capture and experiment with my own spontaneous musical ideas. Towards this end, I am building Band-OUT-of-a-Box (BoB), a software system that interacts with a live, improvising (monophonic) musician in the jazz/blues se tting. My goal is for BoB to provide personalized improvisational companionship to a specific musician, trading “musically-appropriate” short improvised solos on top of a fixed harmonic background. Spontaneity and personalization make it crucial that the system automatically configures itself,learning its aesthetic musical sense from its user’s improvisational example. As such, my focus is on developing techniques that enable BoB to perceive and generate solos in a musically-appropriate, musician-specific manner. Here I will focus on the perceptual part, which provides the abstraction needed to guide the musically-intentful generation discussed in (Thom 2000b).