A method and a device are set forth which make it possible to set an accuracy in a voice recognition process. To this end, system parameters of the voice recognition system are determined using an input quantity via a mapping specification, and the voice recognition system is set corresponding to the obtained values. Optionally, a voice recognition system can be adapted during operation, in order to guarantee an application-specific adaptation in a range between maximal quality in the voice recognition and optimally high speed in the execution of the voice recognition. The computer can also be set automatically in that a performance index of the computer has been determined in advance using a program provided for this purpose, which index represents a measure for the input quantity and thus guarantees an adequate operation of the voice recognition system on this computer.