POLE-ZERO DECOMPOSITION OF SPEECH SPECTRA
Original Publication Date: 1979-Jan-31
Included in the Prior Art Database: 2007-Mar-28
Software Patent Institute
Yegnanarayana, B.: AUTHOR [+2]
POLE-ZERO DECOMPOSITION SPEECH SPECTRA B.Yegnanarayana
POLE-ZERO DECOMPOSITION SPEECH SPECTRA
Department of Computer Science Carnegie-Mellon University
Pittsburgh, PA 15213.
This research was sponsored by the Defence Advanced Research Projects ~ g e n c ~
I ARPA Order No. 3597, and monitored by the Air Force Avionics Laboratory under Contract
The views and the conclusions contained in this document are those of the author and should not be interpreted as representing the official policies, either expressed or implied, of the Defence Advanced Research Projects Agency or the U.S. Government.
A new methid for determining the parameters of a pole-zero model for speech spectra is
proposed in this paper. In this method the cepstral coefficients of a signal are split Into two parts, one corresponding to poles and the other to zeros. The decomposition is achieved by
using the properties of the derivative of phase spectra of minimum phase signals. Parameters of the model are derived recursively from the cepstrai coefficients for poles and zeros separately. Since poles and zeros are treated alike and derived independently, there is
no effect of one on the other. The method is illustrated with several examples of speech
spectra. It is shown that in all cases the envelope fit is equally good at peaks as weil as at
valleys in the spectrum. Results of this paper suggest a method of obtaining a linear system model for a given signal using a criterion different from the conventional minimization of mean squared error criterion. Although the method is described for minimum phase signals only, extension of the method to mixed phase signals is trivial, since a mixed phase signal can be split into minimum and maximum phase components using complex cepstrum.
An important problem in signal analysis is the estimation of parameters of a pole-zero model for a given signal spectrum. In this paper we present a general and effective method for determining the parmeters. The method involves separating the effects of poles and zeros based on the properties of the derivative of linear prediction phase spectra reported
8 - recently by the author [I$ Besides the effectiveness of the derivative of phase spectrum, the inherent advantages of linear prediction (LP) and homomorphic filtering approaches are exploited to derive the parameters of the model in a simple and elegant manner. The proposed technique yields a linear system model for a signal using a criterion different from the conventional minimization of mean sqaured error criterion. Although...