TThe aspects of phoneme-based recognition are analysed for both PC and DSP. 
The benefits of context-dependent models are spotted. The description 
of knowledge and data base speech for speech recognition system evaluation is given. 
Experimental results are discussed.