Quasi-periodic signal analysis using harmonic transform with application to voiced speech processing

A new method for quasi-periodic signal analysis and modelling with application to speech processing is proposed in this paper. The model of quasi-periodic signal treats the signal as a composition the two components. Periodic component is modelled as a sum of harmonically related sinusoidal partials with slowly time-varying amplitudes and frequencies, aperiodic component is modulated white noise. Fundamental frequency, amplitudes and phases of the harmonics are estimated in spectral domain using the harmonic spectrum provided by Harmonic Transform (HT). Kernel of the HT is synchronised with fundamental frequency changes providing pulse-train spectrum.