In this paper we classify and analyze dialogue scenarios in automated human-machine systems with speech recognition. We take into account how to identify and correct recognition errors. Also, we evaluate and compare the scenarios according to how long the dialogue lasts, provided that the dialogue completes successfully with at least some given probability. Based on the results of our work, we formulated recommendations for developers of speech dialogue systems to choose a dialogue scenario; the actual scenario depends on how reliable the used speech blocks are.