AUT System for SITW Speaker Recognition Challenge

This document intends to present AUT speaker recognition system submitted to SITW (Speakers in the Wild) speaker recognition challenge. This challenge aims to provide real world data across a wide range of acoustic and environmental conditions in the context of automatic speaker recognition so as to facilitate the development of new algorithms. The presented system is based on the state-of-the-art i-vector/PLDA and source normalization techniques. The system has been developed on publically available databases and evaluated on the data provided by SITW challenge. Taking advantage of the challenge development data, our experiments indicate that source normalization can help speaker recognition system to better adapt to the evaluation condition. Post evaluation analysis is conducted on the conditions of SITW database.