Neural-Fuzzy System을 이용한 한국어 단어인식에 관한 연구 = Experiment for Korean-Word Recongnition Using Neural-Fuzzy System

다국어 초록 (Multilingual Abstract)

This dissertation aims about Korean-word recognition using the neural networks and fuzzy inference system (NFS), which is absorbing fluctuation due to personal difference in order to develop recognition rate. This method regards propability index observation each word model as input parameters of neural networks, and recognizes input words using fuzzy ingerence with membership function with is consisted with output of neural networks. Word recognition using fuzzy inference to develop recognition rate, which is absorbing flucturation due to personal difference, speaker jindependent word recognition system of a cause false-recongition.
To evaluate validity of this method, this experiment has been carried out female and males speech data that is made 28DDO local area names. As a result of testing HMM (DHMM) with 8 state, condeword is 64, the recongition rate 91[%] , as a result of testing neural network with 64 codeword the recognition rate is 89 [%]. Finally, as a result of testing NFS with 64 codeword which is in the best condition in former tests, the recongition rate is 96 [%]