Využití extrakčních metod pro zpracování řečových signálů

Abstract

This master's thesis addresses the issues of speech signal processing and extraction using various types of computational methods. The theoretical section contains an overview of the characteristics of human speech, processing techniques, and methods for signal extraction. The following experimental part describes the design of a custom solution, the creation of relevant datasets using an application in the LabVIEW environment, and the resulting data processing software. This application enables data loading, visualization, processing, and subsequent saving of the results. The filtering effectiveness is defined both by metrics describing the signal and noise characteristics and by methods evaluating the resulting sentences after speech-to-text conversion. The conclusion of the thesis is devoted to a detailed analysis of the obtained data, evaluation of applied algorithm effectiveness on both synthetic and natural datasets, discussion about the limitations of the proposed solution, and future improvements.