The Influence of Daubechies Wavelet Order on Speech Recognition

Abstract

This paper will present the results of speech recognition based on different Daubechies wavelet orders. Two speakers (one female and one male) were analyzed in two speech modes: normal and whisper. The patterns are used from the Whi-Spe database. As an input to the recognition system, the Daubechies wavelet feature vectors with different orders were used. As a back-end of the system, the standard Dynamic Time Warping method was considered. The results are given in the form of tables and histograms. They suggest which order of Daubechies is the most convenient for this kind of speech recognition.