Time-correlation analysis of nonstationary signals with application to speech processing

Abstract

This paper proposes a new method of displaying and analyzing the evolutionary correlation structure of nonstationary signals. The method, called time-correlation analysis (TCA), is based on a filter-bank approach for stochastic signal characterization known as parametric filtering. Some properties of the TCA method are discussed that can be used to interpret the TCA plot. Examples of an application to speech analysis are given.