Speaker verification method based on combination of auto-associative neural network and Gaussian mixture model-universal background model

Abstract

The invention discloses a speaker verification method based on combination of an auto-associative neural network (AANN) and a Gaussian mixture model-universal background model (GMM-UBM), which can improve the performance of a speaker verification system. The invention has the following advantages and effects: the method takes the advantages of the AANN and the GMM into full account, the AANN is embedded into the GMM-UBM, a two-stage learning method is put forward, the parameters of the GMM and the AANN are alternately updated, the maximum likelihood probability is used as the common target for training the GMM and the AANN, thus the AANN can learn the difference among eigenvectors and map the eigenvector set to subspace which increases the likelihood probability, and the learning characteristics of the neural network can further eliminate the mismatch effect of the channel. Experiments show that the speaker verification method can effectively reduce the error recognition rate of the system.

Description

Speaker Verification Method Based on Combination of Auto-associative Neural Network and Gaussian Mixture Background Model technical field The invention relates to a speaker confirmation method, in particular to a speaker confirmation method based on the combination of an auto-association neural network and a Gaussian mixture background model. Background technique Automatic speaker verification, especially text-independent speaker verification, is playing an increasingly important role in areas such as access control, credit card transactions, and courtroom evidence, with the goal of confirming that the speaker under test is who he claims to be . In the speaker confirmation method, the method based on the Gaussian mixture background model (GMM-UBM) has been paid more and more attention. Because of its high recognition rate, simple training, and small training data requirements, it has become the mainstream recognition method at present. . The support vector machine (SVM)...

AI Technical Summary

Problems solved by technology

It is foreseeable that if the neural network technology and the Gaussian mixture background model are combined, the recognition rate of the system will be greatly improved, but there is no such method yet

Images