A Segmental Spectral Flatness Measure for Harmonic-Percussive Discrimination

In a variety of applications of audio signal processing, for example blind source separation (BSS), harmonic signals need to be treated differently from percussive signals. Thus, recognizing if a signal is harmonic or percussive is a very helpful and frequently used preprocessing step. Different measures have been proposed to capture either the harmonicity or the percussivity of a signal, using knowledge of the spectral shape of the signal or investigating its periodic behavior. In this paper, we propose a new method for distinguishing harmonic and percussive signals and evaluate it in a BSS environment. We compare it to three other state-of-the-art measures. Our experiments show, that our method obtains better results than the traditionally used methods.