Research On Algorithms For Speaker Recognition

Speaker recognition is a kind of identity authentication method based on speech.In order to speed up the application of speaker recognition in real business,it is of great significance for the research and implementation of speaker recognition technology.Text-independent speaker verification is one of the main areas of speaker recognition.The state of art speaker verification system are mostly based on probability statistics model.In the case of sufficient data,the Gaussian Mixture Model-Universal Background Model(GMM-UBM)has a good performance.However,due to the influence of background noise and channel mismatch,it's difficult to further improve the recognition performance.Total variation factor analysis(I-vector)technology mapping varying lengths to low dimensional vector solves the problem.Linear Discriminant Analysis(LDA)and Probabilistic Linear Discriminant Analysis(PLDA)are common channel compensation techniques,but the latter is often used as a scoring tool.Based on the GMM-UBM model,this paper studies the speaker verification system based on I-vector and PLDA model.The main contents of this paper are as follows:(1)In this paper,a framework of speaker recognition system based on cloud platform is proposed.This paper analyzes the process of speech preprocessing and the extraction process of Mel-frequency cepstral coefficient(MFCC)features based on human auditory perception.(2)A speaker recognition system based on GMM-UBM model is constructed.The training process of UBM model and MAP adaptive matching process are introduced in detail.By setting up experimental database,to investigate the system effects on the number of speakers for UBM training,the number of Gauss components in the model,the length of the training speech and the dimension of the MFCC feature.(3)Constructed another speaker recognition system based on I-vector and PLDA model.Analyzed I-vector extraction algorithm and PLDA model,the performance of different systems are compared,the effects of norm transform,I-vector feature dimension and PLDA factor dimension on the performance of the system are discussed.(4)LDA and WCCN are used to compensate channel noise and reduce the dimensionality of i-vector,and the effect of the technique on the experimental results is analyzed.To improve the classification performance of LDA,an improved classification algorithm is proposed and verified by experiments.