The SpeakIn System for VoxCeleb Speaker Recognition Challange 2021

TL;DR

This paper presents a speaker recognition system for VoxCeleb Challenge 2021 that combines multiple models and techniques, achieving top performance with a fusion of nine models trained solely on VoxCeleb2-dev data.

Contribution

The report introduces a multi-model fusion approach with domain-based fine-tuning and back-end refinement, leading to state-of-the-art results in VoxCeleb speaker recognition.

Findings

Achieved first place in VoxSRC 2021 tracks 1 and 2.

Attained a minDCF of 0.1034 and EER of 1.8460%.

Demonstrated effectiveness of model fusion and domain adaptation techniques.

Abstract

This report describes our submission to the track 1 and track 2 of the VoxCeleb Speaker Recognition Challenge 2021 (VoxSRC 2021). Both track 1 and track 2 share the same speaker verification system, which only uses VoxCeleb2-dev as our training set. This report explores several parts, including data augmentation, network structures, domain-based large margin fine-tuning, and back-end refinement. Our system is a fusion of 9 models and achieves first place in these two tracks of VoxSRC 2021. The minDCF of our submission is 0.1034, and the corresponding EER is 1.8460%.