The paper presents the results obtained using a large vocabulary continuous speech recognition system for the Serbian language, based on the open-source Kaldi speech recognition toolkit. Data preparation procedures are described in brief, giving special attention to the particularities of the Serbian language. The original, proposed recipes were modified based on our requirements and they are presented here in detail. The results are provided for the system comprising 3000 regression tree leaves and 25000 Gaussians, with a test vocabulary of more than 14000 words and using a trigram-based language model. The acoustic models were trained using a database of about 90 hours of speech (20000 utterances). Word recognition accuracy of approximately 98% is reported.