Дослідження ефективності вейвлет-аналізу мовних сигналів для розпізнавання мовців згортальними нейронними мережами

Abstract

Master’s qualification work is devoted to the study of the effectiveness of wavelet analysis of speech signals for the speaker’s recognition by convolutional neural networks. The work covers existing methods of calculating primary characteristics of the speech signals, which using for speaker identification. Presented classification of neural networks, considered algorithms for functioning speaker’s identification system. Were developed the algorithms of the pre-processing the speech signal, the allocation of informative features and the classification process for the speech recognition system, also were designed the USE-CASE diagram, UML-diagrams of the activity, sequence and statechart. Also, were studied the effectiveness of wavelet analysis of speech signals for the speaker’s recognition.