Wed-1-7-2 THUEE System for NIST SRE19 CTS Challenge

Abstract: In this paper we present the systems submitted by THUEE to NIST 2019 Speaker Recognition Evaluation CTS Challenge (SRE19). Similar to the previous SREs, mismatch of language and channel between the training sets and enrollment/test sets, remains as a major challenge in this evaluation. Our primary system was a linear fusion of six subsystems, which focus on: (i) building deeper and wider speaker embedding model trained on a great number of public large-scale data resources, (ii) exploring novel speaker discriminative embedding system combined with phonetic information, and (iii) investigating effective back-ends against the domain mismatch. The final resulting systems are shown to be both complementary and effective in achieving overall improvements of speaker recognition performance.