Here are a number of .mp3 sound examples of the beat tracking algorithm from the paper "Beat Tracking of Musical Performances Using Low-Level Audio Signals" by Bill and James Sethares and Robin Morris. This website will be expanded later on to provide an introdution to the algorithm, but for now it contains examples so that you can hear the algorithm in action.
The output of the beat tracking algorithm is a sequence of times that are intended to represent when 'beats' occur: when listeners 'tap their feet'. To make this accessible, an audible burst of noise is superimposed over the music at the predicted time of each beat. By listening, it is clear when the algorithm has 'caught the beat' and when it has failed.
The particle filter was applied to a variety of different pieces in different musical styles including pop music ('Norwegian Wood' by the Beatles, 'Mr Tambourine Man' by the Byrds), jazz ('Take Five' by David Brubek), classical (Scarlatti's Sonata K517 in D minor, 'Water Music' by Handel), film ('Theme from James Bond'), folk ('The Boxer' by Simon and Garfunkel), country ('Angry Young Man' by Steve Earl), and bluegrass ('Man of Constant Sorrow'). In all of these cases the algorithm located a regular lattice of times that correspond to times that a listener might tap the foot. Each .mp3 is about thirty seconds long.
There is also a second version of the 'Theme from James Bond' where the beat was identified at twice the normal speed. In this case, it's not a mistake that a person would be likely to make because it's kind of frenetic, but hey, it's just an algorithm.
There are three versions of 'Lion Says' by Prince Buster and The Ska. In one, the algorithm taps with the on-beat, in the next, it taps with the off-beat, and in the third, it taps at twice the nominal rate.
Finally, there is 'Pieces of Africa' by the Kronos Quartet, which is discussed in detail in the paper.
To get to my homepage, click here.