Optimization of Speaker Verification using Adapted Gaussian Mixture Models for High Quality Databases

R.S. Kandovan, M.R.K. Lashkari, and Y. Etemad (Iran)


GMM, UBM, speaker verification.


In This paper, speaker Verification is simulated using Adapted Gaussian Mixture Model-Universal back ground Model (GMM-UBM) [1]. The voice resource used during this research is supplied by a Tv-card used in computers and has high SNR without band-width limitation. The results of this research are closer to the optimum state and have differences from other results obtained in other researches because of the following two reasons. First, high quality of voice resource. Second, use of all parameters existing in UBM, for achieving a speaker model. For example results achieved in this paper disproof claimant of Mr. Raynolds in[1].

Important Links:

Go Back