A speech driven word-processor has been devised which is based on Palantype machine shorthand transcription. Although the system uses a trained human operator, it does provide a very high performance simulation of a speech recognition system. This particular simulation does not suffer from any unrealistic restrictions on speed of dictation, such as occur with typewriter-keyboard based simulations. The simulation has been set in the context of a 'speech-driven' office, This simulation, and the associated experiments, are being used to produce a realistic assessment of users' response to such a system, and measures of performance criteria required. It is also enabling the development of appropriate dialogue design strategies for speech operation of word processors.