Research Of Feature Extraction For Voice Based On Wavelet Transform

Speaker recognition is a new identification technical that based on speaker voice reflected and behavioral characteristics. And the difference between speech recognition is that it does not pay attention to is included in the speech signal into a text messages well as the semantic content of information. But rather focus on is included in the speech signal in the personal characteristics. Extract the speaker's characteristics such personal information in order to order to achieve recognition speaker purpose. it's have many features, such as voice is a person's inherent characteristics, will not be lost or forgotten, facilitate collection and low cost system equipment it is precisely because the medium of their own voice special advantage, so that the speaker recognition is rapidly becoming practical, can be widely applied in all walks of life. This paper is mainly about the speaker recognition system based on formant that using wavelet packet transform. And use Gaussian mixture model (GMM) technology and text-dependent speaker recognition.Mainly works in this paper:Firstly,In the feature extraction part, proposed a kind of method to distinguish the formant using wavelet packet transform (FDWPT).the basic idea of algorithms of speech signal, Firstly the method applies wavelet packet transform to the speech signal. So a wavelet decomposing value is gotten and then combined with the formant features, select the appropriate node to calculate the energy. And be normalized as an identification characteristic parameters.Secondly,Using Gaussian mixture model (GMM) for identification, in the previous study, based on the parameters of the formant as a feature for Gaussian mixture model to identify. In the end, introduced the system and various experiments, the results of the analysis and comparison of various parameters on the experimental settings are verified, the future prospect of research work carried out.