This paper describes a tempo induction and beat tracking system based on the efficient strategy (initially introduced in the BeatRoot system [Dixon S., "Automatic extraction of tempo and beat from expressive performances." Journal of New Music Research, 30(1):39-58, 2001]) of competing agents processing musical input sequentially and considering parallel hypotheses regarding tempo and beats. In this paper, we propose to extend this strategy to the causal processing of continuous input data. The main reasons for this are threefold: providing more robustness to potentially noisy input data, permitting the parallel consideration of a number of low-level frame-based features as input, and opening the way to real-time uses of the system (as e.g. for a mobile robotic platform). The system is implemented in C++, permitting faster than real-time processing of audio data. It is integrated in the MARSYAS framework, and is therefore available under GPL for users and/or researchers. Detailed evaluation of the causal and non-causal versions of the system on common benchmark datasets show performances reaching those of state-of-the-art beat trackers. We propose a series of lines for future work based on careful analysis of the results.
|Name||Proceedings of the 11th International Society for Music Information Retrieval Conference, ISMIR 2010|
|Conference||11th International Society for Music Information Retrieval Conference, ISMIR 2010|
|Period||9/08/10 → 13/08/10|