The presented paper describes a new developed segmentation method for the extraction of phonetic segments from Czech speech signal The use of the proposed method is intended for the machine recognition of short continuous spoken Czech sentences. As the Czech language is a Slavonic language containing some original phonemes, a brief definition of the complete phoneme set and a short description of Czech speech phonetic analysis is presented in the second paragraph. The foundations of the developed segmentation method based on the speech signal cepstrum parameters evaluation and the experimental results obtained on the small database of continuous speech consisting of short Czech sentences spoken by male speakers are presented in the main part of the paper. Additionaly, the comparison of the developed simple method with the phoneme recognition procedure based on the application of classic dynamic programming method (DTW function) and on the using HMMs is discussed in the last paragraph. The primary acoustic analysis of the Czech speech signal and the cepstral parameters evaluation have been realized either by means of the "Soundblaster Pro" special PC interface card or by means of the "SUN-OpenWindows Audio-Tool".

Keywords: Speech Signal Preprocessing, Feature Selection and Extraction, Acoustic-Phonetic Decoding, Phoneme Segmentation, Phoneme Classification, Word Hypothesis Generation