Publication Details
Analysis of the BUT Diarization System for Voxconverse Challenge
Glembek Ondřej, Ing., Ph.D.
Matějka Pavel, Ing., Ph.D.
Rohdin Johan Andréas, M.Sc., Ph.D. (DCGM)
Burget Lukáš, doc. Ing., Ph.D. (DCGM)
Diez Sánchez Mireia, M.Sc., Ph.D. (DCGM)
Silnova Anna, M.Sc., Ph.D. (DCGM)
Speaker Diarization, Variational Bayes, HMM,VoxConverse, VoxSRC Challenge
This paper describes the system developed by the BUT team for thefourth track of the VoxCeleb Speaker Recognition Challenge, focusingon diarization on the VoxConverse dataset. The system consistsof signal pre-processing, voice activity detection, speaker embeddingextraction, an initial agglomerative hierarchical clusteringfollowed by diarization using a Bayesian hidden Markov model, areclustering step based on per-speaker global embeddings and overlappedspeech detection and handling. We provide comparisons foreach of the steps and share the implementation of the most relevantmodules of our system. Our system scored second in the challengein terms of the primary metric (diarization error rate) and first accordingto the secondary metric (Jaccard error rate).
@inproceedings{BUT175790,
author="Federico Nicolás {Landini} and Ondřej {Glembek} and Pavel {Matějka} and Johan Andréas {Rohdin} and Lukáš {Burget} and Mireia {Diez Sánchez} and Anna {Silnova}",
title="Analysis of the BUT Diarization System for Voxconverse Challenge",
booktitle="ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)",
year="2021",
pages="5819--5823",
publisher="IEEE Signal Processing Society",
address="Toronto, Ontario",
doi="10.1109/ICASSP39728.2021.9414315",
isbn="978-1-7281-7605-5",
url="https://ieeexplore.ieee.org/document/9414315"
}