Research On The Method Of Confirmation Of Speaker Identity Based On Voiceprint Recognition

Voiceprint recognition technology is a biometric authentication method. It extracts the speaker identity information from each speaker’s speech parameters.Voiceprint recognition is widely used in many applications with its flexible,economical and practical advantages.This paper focuses on the text-independent, short time text-dependent voiceprint recognition and several related applications. For text-independent voiceprint recognition, Gauss mixture model( GMM) and Gaussian Mixture Model Universal-Background Model( GMM-UBM) based systems have been focused, in which the details of model structure, model parameter estimation have been reconsidered.Furthermore, a software with graphical voiceprint recognition user interface has been designed based on the text-independent GMM-UBM architecture. Based on this software, the training of relevant model parameters and the matching recognition process have been implemented, and how the number of GMM components affects the recognition performance has also been examined and analyzed through experiments.In the study of text-dependent voiceprint recognition, both the speaker identity and the text embedded in the speech should be examined. This thesis proposes a new kind of text-dependent speaker recognition method by integrating a speech recognition block into the voiceprint system. Compared with the traditional GMM-UBM or DTW approaches, the proposed method can result in better speaker identity and text recognition performances, especially in short-time voiceprint recognition conditions. Experimental results show that, the proposed method can reduce a relative 8.9% false alarm error while keeping the same missing alarm error.Finally, based on the GMM-UBM voiceprint recognition framework, two new application ideas have been explored. The first one is the file folder multi-layer encryption system, the innovation of this system is that, both the voiceprint and traditional encryption technology are applied to protect the security of the file folders.The second one is a application software in which the mobile phone bank transfer with voiceprint recognition is implemented. User’s voiceprint information is taken as a bank transfer payment password during their bank business. That’s to say, the voiceprint recognition technology is taken as a new protection service by banks, Both of the above two voiceprint applications may bring a solid foundation for promotingthe wide application of voiceprint recognition technology.