A speech recognition system for the Polish language is described. The presentation will focus on an adjustment of the Kaldi toolkit for Polish, our own grapheme to phoneme conversion tool and a corpus of Polish we collected. The approaches to commercial applications will also be described.