COMPARISON OF JOINT TIME-FREQUENCY DISTRIBUTIONS FOR SPEECH SIGNALS.

Abstract

Various bilinear time-frequency distributions are compared for both speech and simple signals. Results for the Wigner-Ville, Margenau-Hill-Rihaczek, and Page distributions are presented. All distributions have been found to be difficult to interpret, and the authors have been able to pinpoint the reasons why spectral features of interest are obscured. For the Wigner distribution, the main difficulty arises because it is not necessarily zero when the signal is zero and also because features in the distribution at a particular time are often greatly influenced by the signal at earlier and later times. For the Margenau-Hill-Rihaczek distribution, the fact that the entire frequency spectrum affects the distribution at any one particular point in time makes it difficult to localize features of interest. The difficulty with the Page distribution is similar to that of Margenau-Hill-Rihaczek except that the running Fourier spectrum influences the distribution at a particular time. Simple examples are used to illustrate these features.