Musical tempo octave error reducing based on the statistics of tempogram

This paper proposes the innovative features of tempogram for the selection of predominant tempo in a two-stage tempo estimation system. At stage one, a tempo-pair estimator identifies the dominant tempo pair from a given audio music. At stage two, the statistical features called tempogram shape vector (tsv) discriminates the predominant tempo from the identified tempo pair. Our experiments demonstrate that the performance, reducing octave error, of the tempo estimation is superior to the previously published results and the proposed two-stage scheme is unique from the others.