The DEFINITION OF INTERPHONEME TRANSITIONS OF UZBEK WORDS BY DISCRETE WAVELET TRANSFORMS

Abstract

The article refers to the field of information processing theory on the example of evaluating the use of discrete wavelet transform for processing and recognition of speech signals. The difficulties of recognizing speech signals are described and the advantages of discrete wavelet transform over other traditional methods of signal processing, such as Fourier transform, are given. An algorithm based on a discrete wavelet transform is defined and described in detail. The speech signal is processed on the basis of a discrete wavelet transform. A detailed algorithm for finding the most informative level (scale), which is one of the basic parameters of the discrete wavelet transform, is presented. For the analysis, a speech signal with a sampling frequency of 22050 Hz was used, which was decomposed in scales from 3 to 6 and then using an algorithm to identify the informative level, the level (scale) that most closely describes the original signal was selected and determined. When determining the informative level, the type of the mother wavelet is important. In this article, we used the mother wavelet from the Daubechies family, namely the 4th order Daubechies. The second main parameter of the discrete wavelet transform - shift is determined based on practice (16-20ms) and is equal to 256 samples. Also described is a step-by-step algorithm in combination with a discrete wavelet transform, which allows segmenting a speech signal into phonemes, as a result of which it becomes possible to switch from an analog signal to text. Based on the proposed algorithms, the results of the process of segmentation of a speech signal into phonemic units are presented. Based on the results of the work done, a number of conclusions were identified that had not previously been identified due to the traditional Fourier transform used