A causal algorithm for beat-tracking A causal algorithm for beat-tracking

This paper presents a system which can perform automatic beat-tracking in real-time from performed polyphonic music in symbolic format. No voice information is needed. The music can be as well a performance recording containing tempo changes and irregularities, or a score transcription. The needed input format is a midi-like sequence containing at least the onset, duration, and pitch features. In certain cases, the only onsets could also provide good results for some rhythmic structures. The chosen approach is quite similar to the one proposed in (Dixon 2001), however Dixon's method can not be implemented in real time. Another difference is that we make an intensive use of markings, in order to detect salient rhythmic events, which offers the possibility to adapt the behaviour of the algorithm according to the type of music which is analysed. The system was tested on Ragtime pieces and on two beatles songs with satisfying results.