Research On Speaker Recognition System Under The Visual C++ 6.0

Speaker recognition is a kind of speech recognition, it is according to the speaker giving voice, confirmed the process of that who is talking, is also the biological characteristics based on sound as identity authentication based on recognition technology. Therefore, to the pronunciation of each speaker from finding speaker personality differences between, it involves pronunciation organs to personality differences, pronunciation track personality differences between, pronunciation habits personality differences between etc.This paper is mainly to the speaker recognition technology two key parts are studied, the following is this paper mainly do research work:This paper designs a text-independent speaker recognition system. For extracting the feature argument of voice well, it needs take the useful speech section firstly, this paper adopt a endpoint detection that based on short-term energy and zero finite difference method. This paper contrast the endpoint detection result of traditional double gate limit endpoint detection method and short-term energy and zero finite difference method endpoint detection method with the same voice signal. Find from experimental result, short-term energy and zero finite difference method endpoint detection method can detect initial point and end point of voice.Short-term energy and zero finite difference method performance, the calculated amount of subsequent stages will greatly reduce and experimentally compared with Traditional MFCC feature extraction algorithm. Experimental results show that high recognition rate of the system, voice length short, easy collection, calculation speed.This article designs the speaker recognition system is in windows XP system's Visual C++6.0 develops under the platform to realize, Visual C++6.0 provides the high visualization's application procedure development kit and a MFC kind of storehouse, may cause the application procedure development become simple.Experimental results show that because of this paper adopted short-term energy and zero finite difference method endpoint detection method and principal component analysis (PCA) and traditional MFCC feature combination of feature extraction algorithm, Therefore this paper established the speaker recognition system not only has the certain recognition rate, and the calculating speed very quickly, with good real-time process performance.