Detecting Meter in Recorded Music

We use the Discrete Fourier Transform and the Periodicity Transform (Sethares and Staley, 1999) to identify the primary rhythmic content of pieces of popular music. The meter of such pieces is deter-mined by a repeating pattern of accents. In order to use the Fourier and Periodicity Transforms, we must £rst reduce the data by £ltering out pitch. We use the data reduction method proposed by Scheirer (1998), which separates recorded music into bands of pitches, of roughly half an octave each, and ex-tracts the pattern of energy bursts in each band. We found that the meters of the pieces we analyzed were characterized by the relationship between periodic energy bursts made by low and high pitched instruments. We have written MATLAB algorithms that implement these methods. Audio examples are available a