Summary: | 碩士 === 國立中興大學 === 統計學研究所 === 103 === This paper is to investigate the 1391 monosyllable in speaker-dependent system.The hidden of Markov model is a way to account the feature. In each class of the mandarin word, we all use the HMM to estimate the parameter. The advantage is the high recognition. We can iterative estimation of parameter, that we have ability of training database. The database can be better. In this paper, we also use the MFCC to account the feature of monosyllable. We put the mandarin word to HMM, by this way to test the recognition. The test is spilt in two situation, is the known an unknown. The recognition of known is about 90% and recognition unknown is about 72%. In this paper we will against this part to investigate.
|