Wavelet transforms for non-uniform speech recognition

Abstract

An algorithm for nonuniform speech segmentation and its application in speech recognition systems is presented. A method based on the Modulated Gaussian Wavelet Transform based Speech Analyser (MGWTSA) and the subsequent parametrization block is used to transform a uniform signal into a set of nonuniformly separated frames, with the accurate information being fed into a speech recognition system. The algorithm needs a frame characterizing the signal where necessary, trying to reduce the number of frames per signal as much as possible, without an appreciable reduction in the recognition rate of the system.