Research Of Extraction Method To Speech Feature Argument In Speaker Recognition System

Speaker Recognition is that determining who is the speaker by his voice,and also is a precognitive technology that based on voice as the status authentication.For this,it needs individual differences among the speakers from their voice,such as individual differences of speakers' speech organs, individual differences of speakers' pronounce sound tracks and individual differences of speakers' pronounce habits.This paper designs a text-dependent speaker identification system.For extracting the feature argument of voice well,it needs take the clean speech signal after removing noisy firstly,this paper removes environment noisy of speech signal used by wavelet mothed.For taking useful speech section,this paper examines vertex of speech signal used by energy-frequency-value.For manifesting superior of energy-frequency-value method,it gives comparison of the two endpoint examination methods are that double-gate thresh-hold method and energy-frequency-value method. Experiments show that energy-frequency-value method can examine start and end of speech signal. The energy-frequency-value only need adjust a relative threshold value for determining vertex of speech,it has obvious superiorities.This paper used linear prediction Ceptral Coefficients(LPCC),Mel-frequency Ceptral Coefficients(MFCC) and Pitch, this further feature extraction method is proposed,in which effective measures such as weight, differential, combination and filter, are taken to explore those voice characteristics that can be used to distinguish different speakers.And it gives comparison of the two endpoint examination methods are that traditional feature extraction and further feature extraction method.Experiments show that further feature extraction method can heighten system function of recognition.The figure of recognition used pattern matching method-Hidden Markov Model(HMM). Experiments show that precognitive rate of this system is good, speech section is short,taking easily and calculating quickly.It use text-dependent speaker recognition system at occasions are that need greatly safety,not only distinguish passwords,but also ensure safety of system.