"Musica Subtilior" is a project that aims to use generative AI to help unpack the complex process of interpreting graphical musical scores. The individuality of each interpretation of any given graphic score highly differs from performer to performer, and this opened doors to inquiries into the creative process itself, in particular, how improvisation and indeterminacy could lead to a wealth of new musical possibilities beyond the traditional, fixed, prescriptive type of Western music notation.

Current AI Models in music are primarily trained on Western classical music, making them highly adaptable to interpret non-traditional graphical scores. For "Musica Subtilior," the models will be trained on different performances and interpretations of a given set of graphic scores to pave the way for identifying patterns in which attributes of graphic scores — shape, color and spatial formatting — are musically interpreted. This information would provide a valuable understanding of emotive and other innate responses in human musical creativity and serve as a case study looking at the biases of models towards Western notation.

In collaboration with Chris Donahue, Dannenberg Assistant Professor, Carnegie Mellon University. Project supported by CMU's AIxArts Incubator Fund.

More Info.