Lecture notes: [ Ссылка ]
Learning Rhythm from Live Music
I will give a brief overview of my musical accompaniment system, known interchangeably as "Music Plus One" and the "Informatics Philharmonic," which plays a flexible accompaniment that follows a live soloist in a concerto setting. The program "listens" to the live player using an HMM, making decisions based on the filtered distribution of the player's score position. A separate "thread" of the program predicts the times of future musical events using a Kalman-Filter-like model that both incorporates basic musical notions of timing and can adapt in a manner analogous to "rehearsing." In the actual performance a prerecorded orchestra-only recording is resynthesized using phase-vocoding to accommodate the predictions of the timing model. The talk will contain a number of illuminating demonstrations of these components, as well as the complete goal.
See other lectures at Purdue MLSS Playlist: [ Ссылка ]
