We introduce a causal approach to tempo tracking for musical audio signals. Our system is designed towards an eventual real-time implementation; requiring minimal highlevel knowledge of the musical audio. The tempo tracking system is divided into two sections: an onset analysis stage, used to derive a rhythmically meaningful representation from the input audio, followed by a beat matching algorithm using auto- and cross-correlative methods to generate short term predictions of future beats in the audio. The algorithm is evaluated over a range of musical styles by comparing the predicted output to beats tapped by a musician. An investigation is also presented into three rhythmicallycomplex beat tracking problems, where the tempo is not constant. Preliminary results demonstrate good accuracy for this type of system. Keywords – Tempo tracking, beat analysis, onset detection, rhythmic analysis
Matthew E. P. Davies, Mark D. Plumbley