elec.qmul.ac.uk

Off-line beat trackers are often compared to human tap-pers who provide a ground truth against which they can be judged. In order to evaluate a real-time beat tracker, we have taken the paradigm of the ‘Turing Test ’ in which an interrogator is asked to distinguish between human and machine. A drummer plays in succession with an interac-tive accompaniment that has one of three possible tempo-controllers (the beat tracker, a human tapper and a steady-tempo metronome). The test is double-blind since the re-searchers do not know which controller is currently function-ing. All participants are asked to rate the accompaniment and to judge which controller they believe was responsible. This method for evaluation enables the controllers to be contrasted in a more quantifiable way than the subjective testimony we have used in the past to evaluate the system. The results of the experiment suggest that the beat tracker and a human tapper are both distinguishable from a steady-tempo accompaniment and they are preferable according to the ratings given by the participants. Also, the beat tracker and a human tapper are not sufficiently distinguishable by any of the participants in the experiment, which suggests that the system is comparable in performance to a human tapper