Pith. sign in

REVIEW 5 major objections 5 minor 30 references

Enhancing Self-Supervised Speaker Verification Using Similarity-Connected Graphs and GCN

T0 review · 5 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A graph-convolutional edge-pruning step on a KNN similarity graph, followed by weighted common-neighbor re-clustering, produces pseudo-labels with NMI 97.04 and lowers self-supervised speaker-verification EER to 1.57% on Vox-O.

desk verdict Plausible gains in self-supervised speaker verification, but the core clustering step is undefined and label leakage via K tuning undercuts the claim of label-free training. read the letter →

arxiv 2509.04147 v1 pith:I3K4VT2X submitted 2025-09-04 cs.SD

Zhaorui Sun , Yihao Chen , Jialong Wang , Minqiang Xu , Lei Fang , Sian Fang , Lin Liu This is my paper · ORCID
classification cs.SD
keywords speakerverificationself-supervisedlearningDINOgraphconvolutionalnetworkpseudo-labelclusteringsimilarityECAPA-TDNNVoxCeleb2
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to show that the bottleneck in self-supervised speaker verification is not the feature extractor but the clustering step that turns unlabeled speech into training labels. It proposes a three-stage pipeline: DINO-style self-distillation produces speaker embeddings; a graph convolutional network prunes a K-nearest-neighbor similarity graph of those embeddings; and a weighted common-neighbor similarity measure re-clusters the pruned graph to yield higher-quality pseudo-labels. Those pseudo-labels, without any ground-truth speaker labels, then train an ECAPA-TDNN backend. The reported result is an equal error rate of 1.57% on Vox-O, 2.01% on Vox-E, and 3.46% on Vox-H, with pseudo-label NMI of 97.04. If the result holds, it means a major part of self-supervised speaker-verification progress can come from better unsupervised label assignment rather than bigger encoders.

What carries the argument

The load-bearing mechanism is the edge-pruning step on a KNN similarity graph, driven by a GCN trained with a class-based subgraph sampling strategy. Nodes are utterances, edges are K=50 nearest-neighbor connections, and the GCN predicts whether each edge connects the same speaker; a second clustering pass then uses a weighted common-neighbor similarity C(i, j) = sum over common neighbors of alpha_i S(i, k) + alpha_j S(j, k), where the alpha terms downweight nodes with many neighbors. The GCN supplies structural, graph-level information that pairwise cosine or Euclidean similarity alone misses; the common-neighbor weight supplies local structural correction during inference.

What would settle it

On a labeled subset of VoxCeleb2, compute the precision of GCN-retained edges against true speaker identity; if pruning does not increase the fraction of same-speaker edges relative to the original KNN graph, the mechanism is not edge correction. A second check is to rerun the pipeline with the class-selection threshold replaced by a random subset of classes: if NMI and EER gains vanish, the gain is selection, not graph learning.

Watch

Extended reading notes

Core claim

The central claim is that graph structure, not just pairwise feature similarity, carries the information needed to clean up pseudo-labels in self-supervised speaker verification. Starting from embeddings produced by a DINO-style self-distillation encoder on VoxCeleb2, the method builds a KNN similarity graph, trains a GCN to classify edges as same-speaker or different-speaker using a class-sampled subset of pseudo-labeled subgraphs, then prunes low-confidence edges from the full graph. A weighted common-neighbor similarity formula, where two nodes' similarity is reinforced by how many common neighbors they share and how strongly those neighbors attach to each, provides the final clustering.

Load-bearing premise

The method's claimed gain depends on assuming the initial clustering is accurate enough that the one-quarter of classes kept for GCN training represent true speaker relationships; if those classes are biased toward easy-to-cluster speakers, the graph network will reinforce the bias instead of correcting it.

Editorial extensions

If this is right

  • Pseudo-label quality becomes a first-class target: the cleanest gains in self-supervised speaker verification may come from label assignment, not representation learning alone.
  • A backend trained only on pseudo-labels, with no label correction inside the loss, is claimed to reach EER 1.57 on Vox-O, 2.01 on Vox-E, and 3.46 on Vox-H.
  • The framework removes the need to pre-specify the number of clusters, a known instability in K-Means and AHC pipelines.
  • Because the stages can be re-run, training a backend, re-embedding, and re-clustering should improve pseudo-label quality over iterations.
  • The common-neighbor weighting is sensitive to graph density: K=50 and K=80 work well, while K=20 and K=100 degrade performance, so the method has a distinct operating range.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial extension: the same graph-pruning step could be tested on other label-free speech tasks, such as diarization, language identification, or emotion clustering, wherever noisy pseudo-labels limit self-supervised learning.
  • Editorial extension: replacing the GCN with a cheaper structural heuristic, such as common-neighbor Jaccard edge pruning, would test whether the gain comes from graph convolution as such or from the edge-pruning signal; the paper does not run this control.
  • Editorial extension: if edge-pruning is the source of the gain, corrupting the initial clusters, for instance by subsampling classes according to utterance count, should change final EER predictably, separating selection effects from actual graph learning.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The manuscript proposes a three-stage self-supervised speaker verification pipeline: (1) DINO self-distillation on VoxCeleb2 to obtain embeddings, (2) construction of a KNN similarity graph, initial clustering, GCN-based edge pruning and re-clustering with a weighted common-neighbor similarity (Eq. 7), and (3) ECAPA-TDNN backend training on the resulting pseudo-labels. The authors report EERs of 1.57/2.01/3.46 on VoxCeleb1-O/E/H and a pseudo-label NMI of 97.04, claiming improvements over prior self-supervised methods. The paper also contains an ablation of similarity formulas and of the KNN parameter K.

Significance. If the reported numbers are reproducible, the method would be a competitive self-supervised speaker verification result: the VoxCeleb1-O EER of 1.57 is lower than the cited baselines and is in line with recent state-of-the-art self-supervised systems. The underlying ideas—GCN-based edge pruning to denoise pseudo-labels and a weighted common-neighbor similarity—are coherent and worth investigating. The paper also makes a positive effort to compare against several existing methods and to ablate similarity choices. However, the core clustering algorithm is never actually defined, key hyperparameters and thresholds are omitted, and the label-based selection of K is inconsistent with the stated no-label claim. These gaps prevent the central claim from being independently verified or attributed to the proposed components.

major comments (5)
  1. [Section III.C] The 'proposed clustering method' is never specified. The text states 'we apply the proposed clustering method to generate initial pseudo-labels' and later 'we apply the clustering method based on the similarity graph once again,' but no algorithm, pseudocode, or reference is given. The claim that the method 'can automatically determine the optimal number of clusters' is asserted without any procedure for doing so. This omission is load-bearing: without a definition of the clustering algorithm, the entire pipeline, the NMI values, and the downstream EER cannot be reproduced or evaluated.
  2. [Section IV.A and Table II] The KNN neighborhood size K is selected by maximizing NMI against ground-truth speaker labels. Table II reports NMI values for K=20, 50, 80, and 100, and the text says K=50 was chosen because it gives NMI 96.6. This directly contradicts the statement in Section IV.A that 'No accurate label information was used during the entire training process.' Tuning K on ground-truth NMI leaks speaker identity information into the hyperparameter selection and can inflate the final numbers. Please report results with a fixed K or a label-free selection criterion, or explicitly state that labels were used for model selection and adjust the self-supervised claim accordingly.
  3. [Section III.C and Section IV.A] The GCN is trained on pseudo-labels generated by the same clustering it later prunes, using only 'approximately one-quarter of the total data' retained by an unspecified threshold. The manuscript never reports the noise level of these initial pseudo-labels, the GCN's edge-classification accuracy, or the values of n1, n2, the low-similarity threshold, the class-retention threshold, and the GCN hyperparameters (layers, hidden size, learning rate, epochs). As a result, the NMI improvement from 95.4 (Method 3) to 96.6 (Method 4) in Table II cannot be attributed specifically to the GCN; it could be an artifact of the class-selection process or of the weighted similarity in Eq. (7). Please provide the full training recipe and ablations that control for the selection step, e.g., using a random quarter of the data.
  4. [Section IV.A and IV.C] There is an inconsistency in the reported NMI values. The setup says the initial pseudo-labels have NMI 96.64 and 5,998 classes, while Table I reports a final NMI of 97.04 and Table II reports Method 4 with K=50 as NMI 96.6. It is not clear whether 96.64 is the initial clustering or the GCN-refined result, and whether 96.6 and 96.64 are the same quantity from different runs. Please specify the exact stage at which each NMI is measured and report the final number of clusters. Also clarify the speaker-count discrepancy: Section IV.A says VoxCeleb2 contains 5,994 speakers, but later says the real labels have 5,964 classes.
  5. [Section IV.B and Table I] No error bars, multiple seeds, or significance tests are reported. The headline improvement over the closest baseline is 1.57 vs 1.66 EER on VoxCeleb1-O, a difference of 0.09 absolute, which may be within run-to-run variability. Please report the mean and standard deviation over at least several training runs, and make code or a detailed reproducibility package available. Without such evidence, the claimed 'excellent performance' cannot be distinguished from random variation.
minor comments (5)
  1. [Eq. (2)] The cross-entropy formula is written as -Σ_{k=1}^{K} P_t(x_k) log P_s(x'), but the logarithm does not depend on k. It should be -Σ_k P_t(x_k) log P_s(x')_k (with the k-th component inside the log). As written, the equation is mathematically meaningless.
  2. [Eq. (5)] There is a typographical error: '1p didj' should be '1/sqrt(d_i d_j)' or similar. The node-wise update equation also omits the normalization inside the summation in a way that is inconsistent with Eq. (4).
  3. [Section III.A and references] DINO is described as 'Distilled Instance-level Contrastive Learning,' but DINO usually stands for 'self-DIstillation with NO labels.' Moreover, reference [9] is cited as the DINO source, but [9] is a regularized distillation framework for speaker verification, not the original DINO paper. Please clarify which implementation was used and cite the original DINO work.
  4. [Figures and text] There are many formatting issues: 'Fig.1' instead of 'Fig. 1', 'V ox-O' with odd spacing, 'V oxCeleb2', and inconsistent ellipsis. These should be corrected in revision.
  5. [Section VI] The Limitations section is a single sentence about embedding quality. It does not discuss the more substantive limitations of the method, such as sensitivity to the selection threshold, the lack of theoretical guarantees for the automatic cluster-number selection, or the computational cost of the full-graph inference.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the final EER/NMI are measured against external ground-truth labels, so the bootstrapped GCN pseudo-label loop does not force the result by construction.

full rationale

The paper's central claim is that the proposed three-stage pipeline (DINO embeddings, GCN-pruned similarity graph clustering, backend ECAPA-TDNN training) improves self-supervised speaker verification EER to 1.57 on Vox-O. Although the GCN is trained on pseudo-labels generated by the same clustering method it later prunes, this is a self-training/bootstrap loop, not a circular reduction: the final EER and NMI are computed against ground-truth speaker labels, and nothing in the equations (e.g., Eq. (4)-(7)) forces the final NMI of 97.04 or EER of 1.57 to equal the initial pseudo-label NMI of 96.64. The improvement could in principle go the other way, so the central result is not equivalent to its inputs by construction. The paper does contain a self-citation to reference [9] for DINO, which overlaps with the authors, but DINO is an external component and the comparison is empirically falsifiable; the current contribution is the clustering/GCN stage, not DINO itself. The self-citation is therefore not load-bearing for the claimed improvement. A legitimate non-circularity concern is that Table II selects K (number of nearest neighbors) by maximizing NMI computed against real labels, and the paper also states 'No accurate label information was used during the entire training process.' This is a label-leakage/correctness risk, but it is not a circularity: selecting a hyperparameter by a metric does not make the reported result equal to the selection criterion by construction. The limitations section (Section VI) acknowledges dependence on embedding quality, which is an external dependency rather than a circular step. Overall, the derivation is self-contained with respect to the external benchmark evaluation, and the bootstrapping design does not reduce to a fit or a self-citation chain. Score 0.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The paper rests on one externally provided representation learner (DINO), one standard GCN update, and a large number of unstated engineering choices. The KNN neighborhood size K is explicitly tuned on NMI, and the similarity and class-selection thresholds that determine GCN supervision are unspecified. No new physical entity is introduced; the 'proposed clustering method' is an unnamed procedure that functions as a black-box assumption rather than a specified algorithm.

free parameters (5)
  • K (KNN neighborhood size) = 50
    K=50 selected by comparing NMI values in Table II (K=20, 50, 80, 100); this is tuning on the evaluation metric.
  • Low-similarity edge threshold
    Section III.C removes low-similarity connections using a 'predefined threshold' but the value and criterion are not reported.
  • Class retention threshold for GCN training = roughly one-quarter of total data
    Section III.C retains only classes above an unstated threshold, so GCN supervision depends on a hidden selection rule.
  • Class sampling sizes n1 and n2
    Section III.C randomly selects n1 classes and n2 samples per class, but the values are not given; subgraph quality depends on them.
  • GCN hyperparameters (layers, hidden size, learning rate, epochs)
    Not reported anywhere; required to reproduce edge classifier training.
assumptions (5)
  • domain assumption Initial pseudo-labels are clean enough to train the GCN edge classifier.
    Section III.C: 'Given that the clustering method is highly robust, we can select a batch of higher-quality pseudo-labels.' This underpins all GCN supervision.
  • domain assumption Same-speaker edges dominate the KNN graph.
    Section III.D: 'the number of positive samples in the graph is far higher than that of negative samples due to the high similarity between different samples of the same speaker.' No edge-label distribution is measured.
  • domain assumption DINO embeddings are speaker-discriminative in Euclidean space.
    Stage 1 and Stage 2 build the KNN graph with Euclidean distance over DINO embeddings; if false, the entire graph is uninformative.
  • standard math Standard GCN update (Eq. 4-5) and cross-entropy edge training behave as expected.
    Section III.B adopts the canonical normalized-adjacency GCN update; treated as background math.
  • ad hoc to paper The graph re-clustering can automatically determine the number of clusters.
    Section IV.B claims automatic determination of cluster count without naming the clustering algorithm or its criterion; this is asserted, not demonstrated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhancing Self-Supervised Speaker Verification Using Similarity-Connected Graphs and GCN." pith.science (2026). https://pith.science/paper/I3K4VT2X

@misc{pith2026250904147,
  author       = {Pith},
  title        = {Pith review of: Enhancing Self-Supervised Speaker Verification Using Similarity-Connected Graphs and GCN},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/I3K4VT2X}},
  note         = {Machine review of arXiv:2509.04147}
}
read the original abstract

With the continuous development of speech recognition technology, speaker verification (SV) has become an important method for identity authentication. Traditional SV methods rely on handcrafted feature extraction, while deep learning has significantly improved system performance. However, the scarcity of labeled data still limits the widespread application of deep learning in SV. Self-supervised learning, by mining latent information in large unlabeled datasets, enhances model generalization and is a key technology to address this issue. DINO is an efficient self-supervised learning method that generates pseudo-labels from unlabeled speech data through clustering, supporting subsequent training. However, clustering may produce noisy pseudo-labels, which can reduce overall recognition performance. To address this issue, this paper proposes an improved clustering framework based on similarity connection graphs and Graph Convolutional Networks. By leveraging GCNs' ability to model structured data and incorporating relational information between nodes in the similarity connection graph, the clustering process is optimized, improving pseudo-label accuracy and enhancing the robustness and performance of the self-supervised speaker verification system. Experimental results show that this method significantly improves system performance and provides a new approach for self-supervised speaker verification. Index Terms: Speaker Verification, Self-Supervised Learning, DINO, Clustering Algorithm, Graph Convolutional Network, Similarity Connection Graph

Figures

Figures reproduced from arXiv: 2509.04147 by the authors.

Figure 1
Figure 1. Unsupervised Speaker Verification System Architecture This figure illustrates the overall architecture of the unsupervised [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 3
Figure 3. GCN training process: The sampled subgraph is input [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 2
Figure 2. Subgraph sampling illustration: Each differently col [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: This figure illustrates the specific details of the [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

30 extracted references · 29 canonical work pages

  1. [1]

    V oxceleb2: Deep speaker recognition,

    J. S. Chung, A. Nagrani, and A. Zisserman, “V oxceleb2: Deep speaker recognition,” in Interspeech, 2018, pp. 1086–1090

    work page 2018

  2. [2]

    In defence of metric learning for speaker recognition,

    J. S. Chung, J. Huh, S. Mun, M. Lee, H.-S. Heo, S. Choe, C. Ham, S. Jung, B.-J. Lee, and I. Han, “In defence of metric learning for speaker recognition,” in Interspeech, 2020

    work page 2020

  3. [3]

    Self-supervised speaker embeddings,

    T. Stafylakis, J. Rohdin, O. Plchot, P. Mizera, and L. Burget, “Self-supervised speaker embeddings,” in In- terspeech, 2019, pp. 2863–2867

    work page 2019

  4. [4]

    Disentangled speech embeddings using cross-modal self-supervision,

    A. Nagrani, J. S. Chung, S. Albanie, and A. Zisserman, “Disentangled speech embeddings using cross-modal self-supervision,” in ICASSP, 2020, pp. 6829–6833

    work page 2020

  5. [5]

    Seeing voices and hearing voices: Learning discriminative em- beddings using cross-modal self-supervision,

    S.-W. Chung, H.-G. Kang, and J. S. Chung, “Seeing voices and hearing voices: Learning discriminative em- beddings using cross-modal self-supervision,” in Inter- speech, 2020, pp. 3486–3490

    work page 2020

  6. [6]

    An iterative framework for self-supervised deep speaker representation learning,

    D. Cai, W. Wang, and M. Li, “An iterative framework for self-supervised deep speaker representation learning,” in ICASSP, 2021, pp. 6728–6732

    work page 2021

  7. [7]

    Self-supervised speaker recognition with loss- gated learning,

    R. Tao, K. A. Lee, R. K. Das, V . Hautamaki, and H. Li, “Self-supervised speaker recognition with loss- gated learning,” in ICASSP, 2022, pp. 6142–6146

    work page 2022

  8. [8]

    Self-supervised speaker verification with simple siamese network and self-supervised regularization,

    M. Sang, H. Li, F. Liu, A. O. Arnold, and L. Wan, “Self-supervised speaker verification with simple siamese network and self-supervised regularization,” in ICASSP, 2022, pp. 6127–6131

    work page 2022

Show all 30 references
  1. [9]

    Pushing the limits of self-supervised speaker verification using regularized distillation framework,

    Y . Chen, S. Zheng, H. Wang, L. Cheng, and Q. Chen, “Pushing the limits of self-supervised speaker verification using regularized distillation framework,” in ICASSP, 2023

    2023

  2. [10]

    Cluster-gcn: An efficient algorithm for training deep and large graph convolutional networks,

    W.-L. Chiang, X. Liu, S. Si, Y . Li, S. Bengio, and C.-J. Hsieh, “Cluster-gcn: An efficient algorithm for training deep and large graph convolutional networks,” in KDD, 2019

    2019

  3. [11]

    An overview of text-independent speaker recognition: From features to supervectors,

    T. Kinnunen and H. Li, “An overview of text-independent speaker recognition: From features to supervectors,” Speech Communication , vol. 52, pp. 12–40, 2010

    2010

  4. [12]

    Joint application of speech and speaker recognition for automation and security in smart home,

    K. A. Lee, A. Larcher, H. Thai, B. Ma, and H. Li, “Joint application of speech and speaker recognition for automation and security in smart home,” in Interspeech, 2011, pp. 3317–3318

    2011

  5. [13]

    Investigating text- independent speaker verification from practically real- izable system perspective,

    R. K. Das and S. R. M. Prasanna, “Investigating text- independent speaker verification from practically real- izable system perspective,” in APSIPA ASC , 2018, pp. 1483–1487

    2018

  6. [14]

    Language agnostic speaker embedding for cross-lingual personalized speech gener- ation,

    Y . Zhou, X. Tian, and H. Li, “Language agnostic speaker embedding for cross-lingual personalized speech gener- ation,” IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 29, pp. 3427–3439, 2021

    2021

  7. [15]

    Modeling relational data with graph convolutional networks,

    M. Schlichtkrull, T. N. Kipf, P. Bloem, R. V . D. Berg, I. Titov, and M. Welling, “Modeling relational data with graph convolutional networks,” in ESWC, 2018

    2018

  8. [16]

    Emerging properties in self-supervised vision transformers,

    M. Caron, H. Touvron, I. Misra, H. Jegou, J. Mairal, P. Bojanowski, and A. Joulin, “Emerging properties in self-supervised vision transformers,” in ICCV, 2021, pp. 9657–9667

    2021

  9. [17]

    A cookbook of self- supervised learning,

    J. Geiping, Q. Garrido, P. Fernandez, A. Bar, H. Pirsi- avash, Y . LeCun, and M. Goldblum, “A cookbook of self- supervised learning,” arXiv preprint arXiv:2003.00168 , 2020

    2003 arXiv

  10. [18]

    A twofold siamese network for real-time object tracking,

    A. He, C. Luo, X. Tian, and W. Zeng, “A twofold siamese network for real-time object tracking,” in CVPR, 2018, pp. 4834–4843

    2018

  11. [19]

    Nearest neighbor pattern clas- sification,

    T. Cover and P. Hart, “Nearest neighbor pattern clas- sification,” IEEE Transactions on Information Theory , vol. 13, no. 1, pp. 21–27, 1967

    1967

  12. [20]

    Artificial neural networks: A tutorial,

    A. K. Jain, J. Mao, and K. M. Mohiuddin, “Artificial neural networks: A tutorial,” Computer, vol. 29, no. 3, pp. 31–44, 1996

    1996

  13. [21]

    Musan: A music, speech, and noise corpus,

    S. David, C. Guoguo, and P. Daniel, “Musan: A music, speech, and noise corpus,” arXiv preprint arXiv:1510.08484, 2015

    2015 arXiv

  14. [22]

    A study on data augmentation of reverberant speech for robust speech recognition,

    T. Ko, V . Peddinti, D. Povey, M. L. Seltzer, and S. Khu- danpur, “A study on data augmentation of reverberant speech for robust speech recognition,” in ICASSP, 2017

    2017

  15. [23]

    Ecapa-tdnn: Emphasized channel attention, propagation and aggregation in tdnn based speaker verification,

    B. Desplanques, J. Thienpondt, and K. Demuynck, “Ecapa-tdnn: Emphasized channel attention, propagation and aggregation in tdnn based speaker verification,” in Interspeech, 2020

    2020

  16. [24]

    The idlab voxceleb speaker recognition challenge 2020 sys- tem description,

    J. Thienpondt, B. Desplanques, and K. Demuynck, “The idlab voxceleb speaker recognition challenge 2020 sys- tem description,” arXiv preprint arXiv:2010.12468, 2020

    2020 arXiv

  17. [25]

    The jhu submission to voxsrc-21: Track 3,

    J. Cho, J. Villalba, and N. Dehak, “The jhu submission to voxsrc-21: Track 3,” arXiv preprint arXiv:2109.13425, 2021

    2021 arXiv

  18. [26]

    The dku-dukeece system for the self-supervision speaker verification task of the 2021 voxceleb speaker recognition challenge,

    D. Cai and M. Li, “The dku-dukeece system for the self-supervision speaker verification task of the 2021 voxceleb speaker recognition challenge,” arXiv preprint arXiv:2109.02853, 2021

    2021 arXiv

  19. [27]

    Snu-hil system for the voxceleb speaker recognition challenge 2021,

    S. H. Mun, M. H. Han, and N. S. Kim, “Snu-hil system for the voxceleb speaker recognition challenge 2021,” V oxSRC, 2021

    2021

  20. [28]

    Self-supervised speaker recognition with loss- gated learning,

    R. Tao, K. A. Lee, R. K. Das, V . Hautamaki, and H. Li, “Self-supervised speaker recognition with loss- gated learning,” arXiv preprint arXiv:2110.03869 , 2021

    2021 arXiv

  21. [29]

    Self-supervised speaker verification using dynamic loss-gate and label correction,

    H. Bing, C. Zhengyang, and Q. Yanmin, “Self-supervised speaker verification using dynamic loss-gate and label correction,” in Interspeech, 2022

    2022

  22. [30]

    Self-supervised speaker verification with mini-batch prediction correction,

    J. Wang, Z. Fang, and L. He, “Self-supervised speaker verification with mini-batch prediction correction,” Inter- speech, 2024

    2024

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.