| Abstract [eng] |
New algorithm for consonants and vowels differentiation is revealed in this paper. This signifacally fast algorithm calculates a feature based on the sum of energies of a filtered signal. This feature allows determining whether a vowel or a consonant is an exploratory. An experiment is done to make estimations how reliable the feature is. The main aim of this paper is to realize and evaluate by the experiments consonants and vowels differentiation algorithm based on the calculation of pitch and the sum of energies of a filtered signal. The object of this paper is the algorithm of consonants and vowels differentiation in speech signal. To realize the main aim there are several tasks to be resolved: 1. describe the definition of the pitch, pitch detection methods and their possible practical use; 2. evaluate mostly used pitch detection methods; 3. analyze existing vowels differentiation and recognition algorithms; 4. analyze mostly used digital filters; 5. according to analysis results implement an algorithm model; 6. realise information system to calculate desired feature; 7. evaluate feature precision and quickness; For these tasks to be resolved there are several research techniques used. 1. Qualitative techniques: pitch detection methods classification, vowel recognition methods classification, scientific literature analysis and filing and UML for algorithm modeling. 2. Quantitative techniques: performance evaluation and comparison of algorithms using mathematical and statistical evaluation criterions. The Paper consists of three major parts. First part defines major concepts, existing pitch vowel recognition and differentiation methods, digital filters. The Second part describes an algorithm model. The third part describes major and supplementary experiments for an evaluation of created model. The Paper consists of 67 pages, 12 tables and 45 pictures and schemes. |
