A speech processing system has an arbitrary number of speech recognition modules (Ei, i=1 . . . n) and speech output modules (Aj, j-1 . . . m). The modules provided respectively for a particular type of speech recognition or, respectively, speech output are selected and activated and parameterized according to the situation of application, dependent on an input signal (ES), by a device (MA) for selecting the modules, so that the selected modules are applied for speech recognition or, respectively, speech output corresponding to their construction.