The present invention relates to isolated word recognition system uses a voiced sound / unvoiced sound / silence information in order to reduce the amount of calculation of the dynamic pressure times the matching algorithm that is used in the conventional speech recognition. First, the voiced sound / unvoiced sound / and voiced sound / unvoiced sound / silence code words from the input voice through a silence classifier, extract each section boundary information for each code, and voiced configured for the input speech signal from among the reference pattern in the database / unvoiced / silence selects the standard pattern having the same code word with code words. Pattern matching unit outputs a correct recognition result by using a piecewise boundary information of the original speech signal performing a similarity comparison of the selected standard pattern with the input speech signal from the pattern database.; Therefore, there is a significant amount of computation reduction effect over the prior art since the input source primarily to compare only the standard pattern with the code words corresponding to the audio object, and a secondary pattern matching is performed by each section.
展开▼
