Fast implementation of Harmonic Transform

There are several methods of accurate representation of signals with time-varying frequency components (e.g. voiced speech). This paper focuses on the Harmonic Transform (HT) and its fast implementation using interpolation of the signal to approximate pulse-train spectrum. The HT requires a priori knowledge of fundamental frequency change. Estimation of this change has been addressed in several papers using spectral flatness. Interpolation in the fast implementation adds noise to the signal and spectral flatness no longer leads to correct estimation of the fundamental frequency change. This is addressed in the paper using gathered log-spectrum and grid search.