This paper presents a research system of computerized stenographer.  
It makes the text from sound records based on the speech recognition system aided by human. 
Large vocabulary (more than 10K words) continuous speech recognition system for a number 
of speakers is used to process recorded files. Human introduces out-of-vocabulary words 
and repairs errors to produce the final text. Personal phonetic rules are listed and 
used to individualize transcriptions for speakers. To improve system performance 
the retraining process is running to take into account repairs. 
Experimental recognition results are presented.