Publication Details
BUT System Description to SdSV Challenge 2020
Glembek Ondřej, Ing., Ph.D.
Lozano Díez Alicia, Ph.D.
Matějka Pavel, Ing., Ph.D. (DCGM)
Novotný Ondřej, Ing., Ph.D.
Plchot Oldřich, Ing., Ph.D. (DCGM)
Pulugundla Bhargav, M.Sc.
Rohdin Johan Andréas, M.Sc., Ph.D. (DCGM)
Silnova Anna, M.Sc., Ph.D. (DCGM)
Veselý Karel, Ing., Ph.D. (DCGM)
short duration speaker verification, phrasedependent PLDA, phrase recognizer, x-vector, TDNN, ResNet
In this report, we describe the submission of Brno University of Technology (BUT) team to the Short Duration Speaker Verification (SdSV) Challenge 2020. For the text-dependent task, our primary submission consists of a simple linear logistic regression score level fusion of different i-vector and x-vector based systems. Our i-vector systems are based on concatenated MFCC and bottleneck features. For both types of embeddings, we use PLDA backends, showing the success of phrasedependent training of PLDA and its combination with a Gaussian linear classifier phrase recognizer. For the task of textindependent speaker verification, we combine three different xvector systems based on TDNN and ResNet architectures.
@inproceedings{BUT171000,
author="Lukáš {Burget} and Ondřej {Glembek} and Alicia {Lozano Díez} and Pavel {Matějka} and Ondřej {Novotný} and Oldřich {Plchot} and Bhargav {Pulugundla} and Johan Andréas {Rohdin} and Anna {Silnova} and Karel {Veselý}",
title="BUT System Description to SdSV Challenge 2020",
booktitle="Proceedings of Short-duration Speaker Verification Challenge 2020 Workshop",
year="2020",
pages="1--5",
address="Shanghai, on-line event of Interspeech 2020 Conference",
url="https://sdsvc.github.io/2020/descriptions/Team56_Both.pdf"
}