Detail publikace
BUT System Description to VoxCeleb Speaker Recognition Challenge 2019
Wang Shuai
Silnova Anna, M.Sc., Ph.D. (UPGM)
Matějka Pavel, Ing., Ph.D.
Plchot Oldřich, Ing., Ph.D. (UPGM)
VoxCeleb Speaker Recognition Challenge, Deep NeuralNetworks, ResNet, x-vector, PLDA, Cosine distance
In this report, we describe the submission of Brno Universityof Technology (BUT) team to the VoxCeleb Speaker RecognitionChallenge (VoxSRC) 2019. We also provide a brief analysisof different systems on VoxCeleb-1 test sets. Submittedsystems for both Fixed and Open conditions are a fusion of4 Convolutional Neural Network (CNN) topologies. The firstand second networks have ResNet34 topology and use twodimensionalCNNs. The last two networks are one-dimensionalCNN and are based on the x-vector extraction topology. Someof the networks are fine-tuned using additive margin angularsoftmax. Kaldi FBanks and Kaldi PLPs were used as features.The difference between Fixed and Open systems lies in the usedtraining data and fusion strategy. The best systems for Fixedand Open conditions achieved 1.42 % and 1.26 % ERR on thechallenge evaluation set respectively.
@inproceedings{BUT168476,
author="Hossein {Zeinali} and Shuai {Wang} and Anna {Silnova} and Pavel {Matějka} and Oldřich {Plchot}",
title="BUT System Description to VoxCeleb Speaker Recognition Challenge 2019",
booktitle="Proceedings of The VoxCeleb Challange Workshop 2019",
year="2019",
pages="1--4",
address="Graz",
url="https://arxiv.org/abs/1910.12592"
}