PRE-PROCESSING OF SIGNAL IN RECOGNITION OF VOICE COMMANDS BY METHOD OF IMPROVED COMPLETE MULTIPLE DECOMPOSITION TO EMPIRICAL MODES

Abstract

Background. In the recognition of speech signals to work in various spheres of human life, the developer has to solve the problem of processing of speech signals, in particular the problem of non-stationarity. To solve this problem, there are various preprocessing methods and it is necessary to choose the best method. Materials and methods. For the preprocessing of the speech signals the best method was chosen, namely, the improved full multiple decomposition into empirical modes with adaptive noise (IFMDEMAN). We modeled the decomposition of speech signals into components using IFMDEMAN, extracted the most informative component and translated it into the frequency domain using the Fourier transform. Results. As a result of this work, a comparative analysis of the selected components for different teams, as well as a conclusion about the correctness of the choice of the method and the choice of the informative component.

Keywords

• speech signals
• preprocessing
• decomposition into empirical modes
• fourier transform