A method is presented for using a small number of bandpass filters and banks of parallel
comb filters to analyze the tempo of, and extract the beat from, musical signals of arbitrary
polyphonic complexity and containing arbitrary timbres. This analysis is performed causally,
and can be used predictively to guess when beats will occur in the future. Results in a short
validation experiment demonstrate that the performance of the algorithm is similar to the
performance of human listeners in a variety of musical situations. Aspects of the algorithm …