Tyrimas skirtas kalbos signalų kokybinių charakteristikų analizei, įvardinant galimybes, kurių pagalba atliekamas garsų atpažinimas, taip pat atskleidžiant kalbos signalų kriterijus, padedančius atlikti kokybinės charakteristikos analizę. Darbe analizei naudojamas kalbos signalų garsynas, kuriame sukaupta diktorių šnekos įrašai. Tokiu būdu tyrimo tikslas - sukurti kalbos signalų kokybinių charakteristikų analizės algoritmą, kuris galėtų automatiškai nustatyti ar padarytas žodžio įrašas atitinka nustatytus kriterijus. Šiam tikslui pasiekti reikia išanalizuoti esamus segmentavimo algoritmus, išskiriant jų privalumus ir trūkumus, sukurti signalų kokybės nustatymo algoritmą, pasiruošti kalbos signalų kokybės nustatymo tyrimui duomenis, įvertinti kalbos signalų kokybės nustatymo algoritmo rezultatus, apibendrinti tyrimo rezultatus. Tyrimui atlikti remsimės moksline literatūra bei apibendrintais metodais, eksperimento įgyvendinimui – eksperimentinio matavimo bei statistinės analizės metodais. Tyrimui vykdyti ketiname sudaryti kalbos signalų kokybės analizės algoritmą, parenkant kalbos signalų kokybės charakteristikų kriterijus. / Research for the qualitative characteristics of the speech signal analysis, a description of the potential that exists to accomplish the recognition of sounds, including speech detection criteria to facilitate analysis was performed for qualitative characteristics. The paper used in the analysis of speech corpus in which the accumulated announcer speech recordings. Thus, the study objective - to establish the qualitative characteristics of speech analysis algorithm which can automatically determine whether the word record made the qualifying criteria. To achieve this requires the examination of existing segmentation algorithms, highlighting their advantages and disadvantages of a signal quality determination algorithm, speech preparation of quality research data, to evaluate speech quality determination algorithm, the results of the research findings. We base the scientific study of literature and summarized the methods for the implementation of the experiment - the experimental measurement and statistical analysis. The study we intend to carry out a speech quality analysis algorithm, the choice of speech quality performance criteria.

The task of our work is segmentation of a speech signal when having a speech waveform and parameters of the segments. We used dynamic programming approach.

In this work is analyzed the use of digital filter to improve the quality of a speech signal. For this purpose the influence of noises on the understanding of language is inspected and the digital filter design with optimal pole placement is presented. In the process of experimentation the speech signal was recorded by using Praat program, generated noise is made by means of Matlab function randn (n) or the sum two or one cosines. The notated speech signal is mixed with the generated noise. The algorithm based on Matlab program is created to sum up the meanings of speech signal with the corresponding meanings of the noise amplitudes. In order to improve the quality of speech signal mixed with the generated noise, the low pass, notch and strip filters is used, depending on noise, which has been used for connection of noise and a speech signal. The used programs are written down by program Matlab. In the end of the work are show the texts of the algorithms, the text of the created filters and groups of words.