Research On The Text-independent Speaker Recognition System

Due to the characteristics of the voice and the needs of modern society, the researchof speaker recognition technology has a great deal of academic and practical value. Butthe existing speaker recognition technology is far from the hopes of people. it is verydifficult that the technology to be used in real life now.The important research of this paper is text-independent of the speaker recognition,focusing on characteristic parameters selection, the matching algorithm and judgmentmethods, especially the dispose method of off part of the proposed speaker. Then specifythe applicable voice bank and feedback information of system.In the part of feature parameters selection, this paper has a detailed research andanalysis on the pitch, formant and Mel frequency cepstrum coefficient. And the Melcepstrum coefficients is be used to the characteristic parameters of speaker recognition,the pitch and formant are be used to classify the speakers. When identifying, it should bejudged firstly that which class the speaker belongs to, and then make matchingcalculations. The experimental results show that, this method not only shorten the time ofrecognition, but also improve the recognition rate.In the matching calculations, the Euclidean distance is be used. using the algorithm inspeaker recognition, it not have to train strictly people in modeling. it only need to extractthe useful characteristic parameters and the store it as speakers’ models.In the method of judgment, using a sure range instead of specifying one person as afinal recognition results, so this paper’s output has three people. This method caneffectively reduce the mistake that the error person to be accepted or the correct person tobe refused. Using this method in the field of Police and Justice can also greatly reduce thescope of the investigation.This paper was also research the off part of people. Selecting two different speechfrom the voice which is to be confirmed, and calculate this two voices using matchingcalculation. Then using this speech to do the corresponding calculation. then it is to becompared that the results obtained by their own voice and other results, the voice librarywhether involves the speaker or not is be acquired.