Improved Gaussian mixture model and application in speaker recognition

Since mixing degree of the traditional Gaussian mixture model is constant, and it does not conform to the characteristics of speaker feature distribution. In this case, the problems of fitting deficiencies or fitting excessive exist and it will affect the speaker recognition rate. A new algorithm to...

Full description

Saved in:
Bibliographic Details
Published in2016 2nd International Conference on Control, Automation and Robotics (ICCAR) pp. 387 - 390
Main Authors Bao, Lingling, Shen, Xizhong
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.04.2016
Subjects
characteristic parameters
feature distribution
Fitting
fitting characteristics
Gaussian mixture model
GMM
IGMM
Mel frequency cepstral coefficient
Signal processing algorithms
speaker confirmation
Speaker recognition
Training
Online AccessGet full text
ISBN9781467398589
1467398586
DOI10.1109/ICCAR.2016.7486761

Cover

More Information
Summary:Since mixing degree of the traditional Gaussian mixture model is constant, and it does not conform to the characteristics of speaker feature distribution. In this case, the problems of fitting deficiencies or fitting excessive exist and it will affect the speaker recognition rate. A new algorithm to improve the Gaussian mixture model was proposed and it was applied to speaker recognition. The algorithm adaptively adjusted the weight, mean and covariance of the Gaussian component according to distribution characteristics of the speaker's characteristic parameter. It made the improved Gaussian mixture model could better fit distribution features of the characteristic parameters of the speaker. Thus the speaker recognition rate was improved. Experiments showed that the speaker recognition rate of improved Gaussian mixture model was higher than the traditional Gaussian mixture model.
ISBN:9781467398589
1467398586
DOI:10.1109/ICCAR.2016.7486761