Design And Implementation Of Online Open-set Speaker Recognition System

With the development of speaker recognition technology, the I-vector speaker modeling technique has achieved great success and has become the state of the art method in speaker recognition system. On this basis, with a view to the practical application, the speaker's enrollment utterances and the number of enrollment speakers are often changed with the increase of test data. Compared with the traditional off-line speaker recognition technology, the online open-set speaker recognition system has a wider application range. In this view, this paper focuses on the research of speaker recognition on the following aspects:(1)Study on the basic theory and key technologies of I-vector, design and implement the speaker recognition system based on I-vector and compare several schemes in each part of the system;(2) Study on the main problem of speaker recognition system, design and implement the online open-set speaker recognition system;(3) Combine with utterance partition algorithm, update and adaptive methods and purification step to improve the online system performance.And according to the threshold setting and decision-making problems of the online speaker verification system, this paper proposes an effective threshold calculation method and a dual threshold decision mechanism,which improves the update performance of the online speaker verification system;(4) The experiments are carried out on all aspects of the above system, and the overall performances are evaluated and analyzed with different evaluation indexes.