TECHCOMB

Volume 1 : Issue 1

Fourier Analysis-Based Tempo Estimation with Application to Music Accompaniment

Authors : Hua-Tsung Chen,Chien-Peng Ho

image

Abstract:

Melody and rhythm are two essential components in music that melody is a spectral structure while rhythm is a temporal one. As one temporal property, tempo is defined as the rate at which music beats occur, so it is also called beat rate. In this paper, a Fourier analysis-based algorithm is proposed for music tempo estimation and beat locating. First, the original music signal is reduced to a detection function, which is capable of manifesting music beats. By applying the discrete Fourier transform (DFT) to the detection function, an energy spectrum can be generated from the magnitude component to show the tempo components of the music rhythm. Furthermore, the phase component can be utilized to indicate the beat times and predict the upcoming beat. The regularity and periodicity of music beats are obtained and missing beats can also be recovered. Extensive experiments are conducted on a comprehensive testing data set containing various genres of music clips. Encouraging results of both formal objective experiments and subjective listening evaluations show that the proposed tempo estimation system is robust to beat strength and not restricted to specific music genres. On the other hand, an interactive interface is also developed, allowing users to select a favorite instrument to accompany the music based on the obtained tempo.