Empirical Wavelet Transform in Speech Signal Compression Problems

Abstract

In this scientific research it is proposed to apply a modern method of empirical wavelet transform based on the construction of a family of adaptive wavelet functions to increase the efficiency of spectral analysis of speech signals, and their subsequent compression or filtering. If we take as a basis the features of the frequency spectrum of the Fourier speech signal, then the task is equivalent to the construction of a set of band wavelet filters. One way to achieve adaptability is to keep in mind that compact wavelet filter media is directly dependent on, where is the information we need in the spectrum of the speech signal, that is, the larger amplitudes of the Fourier spectrum carry more important information to restore function, and hence for speech intelligibility, and small amplitudes are less important. Indeed, the properties of the function of the internal empirical mode are equivalent to the statement, that the spectrum of this function has a compact carrier and is concentrated around a certain frequency depending on the signal under study, which makes this approach adaptive and increases the efficiency of compression and filtering of speech signals.