Factorization of speech signals parametric spectra using multiplicative linear prediction models

A new approach for a speech signal parametric spectral characterization, based on a new factorization method, is proposed in this paper. The theoretical basis of the factorization method for multimode speech signals spectra is presented in the paper. This method is based on the multiplicative polymodels developed by the authors. We also derived equations for autoregressive coefficients calculation and for parametric power spectrum density of the multiplicative models´ calculation. The expressions for factorization of spectra of the speech signal´s phoneme using the multiplicative models are also included in the paper. The factorization method for spectral estimations is shown in the examples of speech signals parametric spectral estimations.