CRSS systems for the NIST i-Vector Machine Learning Challenge

This paper describes the systems developed by the Center for Robust Speech Systems (CRSS), Univ. of Texas Dallas, for the National Institute of Standards and Technology (NIST) iVector challenge. Since the emphasis of this challenge is on utilizing unlabeled development data, our system development focuses on: 1) unsupervised clustering methods to estimate development data labels; 2) build efficient classifier without clustering method. Our results indicate substantial improvements obtained from incorporating one or more of the aforementioned techniques.