Tempo estimation for acoustic signal of tempo variation music

Abstract

This study proposes a tempo estimation method for acoustic signal of tempo variation music. Previous studies observe the power spectrum of envelope curve for its acoustic signal, where its length must be sixty seconds or longer for the purpose of keeping the tempo resolution as 1 bpm. The method then picked the highest power up as estimated tempo. The music with tempo variation that is occurred within the sixty seconds (for example ten seconds) were, however, hard to estimate. In order to estimate tempo for each time of such tempo variation music, the conventional zero-padding method is carried out when performing FFT. Tempo value is therefore estimated under high accuracy yet the estimation errors have not discussed so far. Accordingly, the optimum ratio of zero-padding was investigated. As a result, the optimized ratio of acoustic signal and 0 in the zero-padding is.125—.875 %. Furthermore, tempo value on each one second is estimated for tempo variation music that has orchestral rubato expression. The estimated accuracy of presented method was evaluated by comparing estimated with hand-labeled tempo.