The HTK large vocabulary speech recognition system has been shown to produce state-of-the-art results on American English data. The system uses decision tree state-clustered mixture-density cross-word triphones and statistical N-gram language modelling. Recently, as part of the EC-funded SQALE project, versions of the system have been developed in several European languages. The paper gives an overview of the HTK speech recognition system with American English baseline results, and then describes the progress made in developing British English, French and German versions. The official SQALE evaluation results are reported for each of these four languages and their relative performance is discussed.