This paper evaluates the performance of a speech recognition system using acoustic models trained on multilingual data. The reason in our case for using data from more than one language is that there may not be enough ...