The Design and Implementation of a Kind of Voiceprint Recognition System Prototype |
|
Author | MengZuoHui |
Tutor | ZhaoHong; WenYingYou |
School | Northeastern University |
Course | Applied Computer Technology |
Keywords | Voiceprint Feature Endpoint Detection Feature model Filter |
CLC | TN912.34 |
Type | Master's thesis |
Year | 2010 |
Downloads | 78 |
Quotes | 0 |
Speech is the most direct and convenient means of realizing the communication between people. So speech recognition has become a key technology to achieve this dream. Recognising who is speaking is a key technology of Speech recognition. Voiceprint Recognition abstracts the speaker’s physiological and behavioral features from speech signal. It can’t be imitated, forged and substituted and there are many ways to get the speech signal. Voiceprint Recognition also has gotten enough development in many areas such as the e-business police bank, military affairs and so on. Speech contains both the speaker’s physical characteristics, namely, differences in innate vocal organs, but also contains the speaker’s behavioral characteristics. Voiceprint Recognition which is used in the recognition system does not concern with the contents of speech signal, but the speaker’s features that it’s biological characteristics.This paper has analyzed and studied the previous Voiceprint Recognition technology. In order to obtain high-quality Voiceprint identification, voice files must first be preprocessed, which is important to detect the endpoint of voice. In the voice recording process, there will be background noise. This paper introduces a new adaptive image de-noising algorithm which can remove background noise from speech effectively and prepares for the later abstract of high-quality voiceprint characteristics. In feature abstract, the paper compares the effect of several characteristics of voiceprint, and finally determine the use MFCC, that is use MFCC feature as a voiceprint feature.As reflected in the MFCC feature, after the speech sub-frame, a frame’s internal characteristics.Through the designing of a filter, our secondary treatment of the abstract voiceprint get a new feature.This voiceprint character that includes the character between voice frame, is also include character among the voice frames. It can be more representative of the characteristics of the speaker. In the identification stage, this paper introduces a kind of self-learning identification algorithm. Meanwhile, the original voiceprint character can be optimized. It solves the problem of the voiceprint feature model getting worse over time.