The paper presents advances in a multi-level automatic speech understanding approach that 
is initially developed for highly inflective languages with relatively free word order. 
The idea to use sub-word units for recognition vocabulary appears productive since word lexicon 
growth leads to practically no sub-word items. 
On the first level a sub-word-based grammar phoneme recognizer is applied, 
which output is post-processed at the second level. 
The described model of post-processing involves acoustic and phonetic features together with lexicon. 
The ways to select a set of sub-word units like syllables are considered. 
A mono-speaker corpus for Ukrainian continuous speech used in experimental research is described.  
Experimental results, problems and future research are discussed.
Title: Phoneme- and syllable-based speech recognition comparison for Ukrainian.
Index items: speech recognition, phoneme, syllable, sub-word units, Ukrainian, continuous speech.