In this paper a universal real-time speech recognition system-RTSRS(01) is described. On the basis of the previous work[1], the parameters of a spoken command are normalized in the time domain. Using the binary spectrum as the final recognition parameters, which can largely reduce the amount of memory necessary for each reference command, and adapting a new method for calculating the separation between two spoken commands to be compared, it is possible to make the system RTSRS(01) capable of identifying single entities in a vocabulary of 200 items in real-time. The results of recognition for a specific speaker are as follows: 10 spoken Chinese digits-99.7%; 20 sentences (7 syllables for each)-99.7%; 100 phrases (4 syllables for each)-99.5%; 150 phrases (4 syllables for each)-99.3%; 200 phrases (4 syllables for each)-98.8%; 400 phrases (4 syllables for each)-97.7%. It has been shown by informal experiments that the system RTSRS(01) can be used to identify the vocabularies which include items with different numbes of syllables; furthermore, for the first 20 English digits and the names of BASIC statements, the correct recognition rate is also high.