The DaimlerChrysler speech recognizer is specialized for robust speech recognition in noisy environments, in particular for command and control applications. The recognizer that is used in cars has fixed grammars, which restrict the speaker to using short commands. This paper presents methods that allow the user to speak more freely and add spontaneous words to the commands: language modelling, confidence measures, and out-of-vocabulary words. Both the grammars and the statistical language models allow a dynamic update of categories (lists of words of given classes). Furthermore, the grammar descriptions can be extended on the fly. The quality of the recognition results is assessed by confidence measures, which can also be used to detect out-of-vocabulary words. Several units of the recognizer can be run in parallel with different lexica and language models. Finally, several methods are presented that achieve robustness of the recognizer towards varying acoustic conditions, for example speaker changes, changes of the language that is spoken, and background noise. An echo-cancellation method gives the functionality that the user may barge-in when the dialogue system makes a speech output, so the speaker's voice is separated out.