Soft frame margin estimation of Gaussian Mixture Models for speaker recognition with sparse training data | IEEE Conference Publication | IEEE Xplore