Yermolenko T. Wavelet-analysis application for speech signals preprocessing in segmentation, classification and phoneme recognition tasks

The thesis purpose is a development of methods and algorithms for speech signals preprocessing and speech recognition system based on phoneme recognition using wavelet-analysis. The object of investigation is speech signal; the subject of investigation is models and methods of digital signal processing and speech recognition. For the first time the system of features for phonemes recognition based on relative wavelet-spectrum energy distribution was proposed; the methods of noise level reduction and speech boundaries detection were advanced by using of broad speech sounds classification; was evolved speaker- and intensity-independent methods of speech signal segmentation able to search inter-phoneme transitions due to analyzing of dynamics of wavelet-spectrum power. The offered speech signal processing and analysis methods are introduced into an intelligent voice interface within the work "Development of information technology speech synthesis and voice information recognition" of Glushkov Institute ofCybernetics NAS Ukraine.