Dipl.-Ing. Dr. Gerhard Widmer

Austrian Research Institute for Artificial Intelligence, and University of Vienna

Artificial Intelligence and Musicology: The Computer Learns Musical Expression

This talk presents current research work at the Austrian Research Institute for Artificial Intelligence dealing with the question, To what extent and under which conditions can intelligent computer programs learn musical performance? In particular, two projects are introduced. Both deal with the development of AI programs which are to learn to play new pieces of music with appropriate expression, based on examples played by musicians. First, a general-level description of the problem, the underlying musicopsychological assumptions, and the AI approaches and learning models used will be given. Following this overview, examples of pieces interpreted by the computer with respect to musical expression will be played, documenting the capabilities of the learning programs as well as their current limits.

Musical expression is an extremely complex cognitive and artistic phenomenon, the details of which are still largely unknown. The research direction introduced in this talk thus must be understood as more than an interesting playground: a well-founded combination of artificial intelligence and musicology can yield interesting new results for both involved fields. This will (hopefully) be shown in this presentation.