A parametric representation of short-time power spectra based on the acoustic properties of the ear

Speech can be decomposed into a source signal and a linear filter. The power spectrum of the filter approximates the power spectrum of the speech signal. The aim of this paper is to relate the power spectrum of the filter (PSF) to the psycho-acoustic properties of the human ear. A perceptual power spectrum (PPS) is defined, which contains only the relevant auditory information of the PSF. On the basis of a specific model of the auditive system an easy to implement mapping from the PSF to the PPS is described. By this procedure the PPS can be regarded as a band-limited signal, which can be reconstructed from its samples. The samples are the parametric representation of the power spectrum.