Automatic Composition from Non-musical Inspiration Sources.
Robert Eugène Smith, Aaron W. Dennis, Dan A. Ventura · 2012
In this paper, we describe a system which creates novel musical compositions inspired by non-musical audio signals. The system processes input audio signals using onset detection and pitch estimation algorithms. Additional musical voices are added to the resulting melody by models of note relationships that are built using machine learning trained with different pieces of music. The system creates interesting compositions, suggesting merit for the idea of computational “inspiration”.