Simple Tempo Models for Real-time Music Tracking

The paper describes a simple but effective method for incorporating automatically learned tempo models into realtime music tracking systems. In particular, instead of training our system with ‘rehearsal data’ by a particular performer, we provide it with many different interpretations of a given piece, possibly by many different performers. During the tracking process the system continuously recombines this information to come up with an accurate tempo hypothesis. We present this approach in the context of a real-time tracking system that is robust to almost arbitrary deviations from the score (e.g. omissions, forward and backward jumps, unexpected repetitions or re-starts) by the live performer.