Towards the use of full covariance models for missing data speaker recognition | IEEE Conference Publication | IEEE Xplore