A Comparative Analysis of Different Wavelet Transformations on Normal and Whisper Speech Recognition

Discrete Wavelet Transformation (DWT) is very useful in tasks of signal processing. In this paper, four different types of this transformation (Daubechies, Symlets, Coiflets, and Biorthogonal) are compared in experiments with speech recognition. Two speech modes are investigated: normal and whisper, and two speakers (one female and one male) from the Whi-Spe database are included. For the back-end of the automatic speech recognition (ASR) system, the standard Dynamic Time Warping method was applied. The results are presented in the form of tables and histograms. They suggest which wavelet transformation is the most convenient for this kind of speech recognition.