Research On Speaker Recognition Method Based On Text Independent

Voiceprint recognition,also known as speaker recognition.Text-independent speaker recognition refers to identifying the speaker’s identity through speech with indeterminate content.It is more practical than text-independent speaker recognition.With more and more application scenarios of speaker recognition technology,it also puts forward higher requirements for the effect of speaker recognition.It is foreseeable that speaker recognition technology will develop in a more user-friendly and convenient direction.For example,users only need to provide a small amount of corpus to quickly complete the identity registration;the police can determine the identity of the suspect through only a few speech.Aiming at the small corpus scene,with the purpose of identifying the speaker,this paper studies a method with high accuracy,fast operation and good recognition performance under the condition of small corpus.The main research contents of this paper are as follows.(1)Aiming at the problem that it is difficult to fully characterize the speaker’s characteristic information with a single characteristic parameter.This paper proposes a combined feature parameter that combines auditory characteristics and vocal tract characteristics,that is,the combined feature parameter of MFCC+Pitch.Using the TIMIT speech library as the dataset,two different recognition methods were used to identify the speakers.Experiments show that,compared with other parameters,the recognition accuracy of the combination parameters proposed in this paper is about 5%-10%higher.It shows that it can more comprehensively characterize the speaker.And it is an ideal parameter choice;(2)In view of the problems of other classification algorithms in speaker recognition,such as long training and recognition time,low accuracy in text-independent recognition tasks,and strong dependence on data.This paper proposes a speaker recognition method based on KNN.KNN is simple in calculation and fair in judgment,using Euclidean distance as a measure.The speech to be tested can be accurately classified.Experimental results show that the recognition accuracy of the proposed method is about 5%higher than other methods.The accuracy is about 10%higher when the training corpus is small.The training time is even less than the comparison algorithm.It shows that the method is accurate and fast in recognition,and performs well under the condition of less training samples;(3)To further improve single-speaker recognition accuracy based on original classification results,fully utilized the frame classification information.This paper proposes an integrated classifications method based on evidence theory for speaker recognition.The KNN with different parameters are used to form a multi-classifier,and then the results of the multi-classifications are fused by D-S evidence theory.The experimental results show that after the fusion of evidence theory,the detection accuracy of single-speaker is greatly improved.It has important research significance.