The VoxCeleb Speaker Recognition Challenge: A Retrospective

TL;DR

This paper retrospectively surveys five years of VoxSRC challenges (2019–2023), detailing two core tasks—speaker verification and diarisation—and four tracks for verification (closed, open, self-supervised, semi-supervised) plus a diarisation track. It documents datasets, challenge mechanics, and winner-method trajectories, and analyzes trends such as the rise of self-supervised and semi-supervised approaches, multilingual testing, and improved diarisation pipelines. Key findings show steady verification gains driven by data augmentation, robust embeddings, and external data, while diarisation benefits from combining VAD, embedding-based clustering, and fusion techniques, with ongoing challenges in overlapping speech and language diversity. The authors distill practical guidance for future organizers and highlight remaining hurdles, including spoofing resilience, noisy/overlapped scenarios, and the need for larger, more diverse, and fair datasets to sustain progress.

Abstract

The VoxCeleb Speaker Recognition Challenges (VoxSRC) were a series of challenges and workshops that ran annually from 2019 to 2023. The challenges primarily evaluated the tasks of speaker recognition and diarisation under various settings including: closed and open training data; as well as supervised, self-supervised, and semi-supervised training for domain adaptation. The challenges also provided publicly available training and evaluation datasets for each task and setting, with new test sets released each year. In this paper, we provide a review of these challenges that covers: what they explored; the methods developed by the challenge participants and how these evolved; and also the current state of the field for speaker verification and diarisation. We chart the progress in performance over the five installments of the challenge on a common evaluation dataset and provide a detailed analysis of how each year's special focus affected participants' performance. This paper is aimed both at researchers who want an overview of the speaker recognition and diarisation field, and also at challenge organisers who want to benefit from the successes and avoid the mistakes of the VoxSRC challenges. We end with a discussion of the current strengths of the field and open challenges. Project page : https://mm.kaist.ac.kr/datasets/voxceleb/voxsrc/workshop.html

Figures (7)

• Figure 1: Timeline showing the progression of the VoxSRC workshops (dots), as well as when key datasets were released (triangles).
• Figure 2: Top winner's performance on VoxSRC 2019 test set. We report the 2nd place's performance for VoxSRC 2020 and 2022, since they are better than the winner on VoxSRC 2019 test set. All other entries are from the 1st place.
• Figure 3: DET curves of the top winners each year on the VoxSRC 2019 test set. The circles are the points where the DCF value is minimum.
• Figure 4: Performance confidence intervals (95%) for the first and second places in each year of the VoxSRC2019 test set. The top figure shows the Track 1 submissions and the bottom figure shows the Track 2 submissions.
• Figure 5: The performances of different models on the pairs only with certain languages in the VoxSRC 2021 test set.