Clustering-based hard negative sampling for supervised contrastive speaker verification

Jakub \.Zak; Konrad Kowalczyk; Mateusz Matuszewski; Micha{\l} Romaniuk; Piotr Masztalski

REVIEW 3 major objections 4 minor 33 references

Clustering-based hard negative sampling for supervised contrastive speaker verification

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Changing only the batch composition improves supervised contrastive speaker verification, cutting EER and minDCF by up to 18 percent relative to strong baselines.

desk verdict Simple, reproducible batch-sampling trick that consistently improves supervised contrastive speaker verification, but the mechanism is under-tested and hyperparameters are chosen on the eval set. read the letter →

arxiv 2507.17540 v1 pith:ATH4FS7I submitted 2025-07-23 eess.AS cs.LGcs.SD

Piotr Masztalski , Micha{l} Romaniuk , Jakub Żak , Mateusz Matuszewski , Konrad Kowalczyk This is my paper · ORCID

classification eess.AScs.LGcs.SD

keywords speakerverificationsupervisedcontrastivelearninghardnegativesamplingclusteringbatchcompositionK-MeansVoxCelebECAPA-TDNN

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 proposes CHNS—clustering-based hard negative sampling—a training-time sampling scheme that improves supervised contrastive speaker verification without changing the loss function or model architecture. The idea is to pre-cluster speakers by their voiceprint centroids, computed from a baseline contrastive model, and then compose training batches so that most negative pairs come from speakers within the same cluster, making them hard negatives. Across lightweight models trained on VoxCeleb2, the method lowers EER and minDCF by up to 18 percent relative to AAMSoftmax, SupCon, and H-SCL baselines, and the gains carry over to CNCeleb(E) and an internal dataset. The central claim is that batch composition alone, not a new loss, drives the improvement.

What carries the argument

The machinery is CHNS, a clustering-based batch sampler. In a pre-processing stage, it builds speaker voiceprints as K-Means clusters of centroid embeddings, justified because squared Euclidean distance on normalized vectors is linearly related to the cosine similarity used at inference. In training, a custom sampler fills a fraction (hard ratio) of each batch from a single cluster, with the rest sampled randomly, so that most negative pairs within the batch are similar-sounding speakers rather than random ones. The loss is the SupCon loss with an optional exponential hardening term $H = e^{\beta s(x,y)}$; CHNS leaves the loss untouched.

What would settle it

Train the same model with CHNS but with K-Means clusters replaced by equally sized random speaker partitions; if VoxCeleb1-H EER stays near the SupCon baseline, clustering is not doing the work. Alternatively, compute the fraction of within-cluster negative pairs that are among the hardest negatives for the final trained model; if that fraction is near chance, the clusters are not encoding confusability.

Watch

Extended reading notes

Core claim

The central claim is that hard negative pairs in supervised contrastive speaker verification are best obtained at the sampler level rather than the loss level. CHNS computes a voiceprint for each speaker by averaging embeddings of 10 utterances from a SupCon baseline, runs K-Means on these voiceprints to form clusters of similar speakers, and fills each batch with speakers drawn from one cluster so that intra-cluster cross-speaker pairs act as hard negatives. With all other training parameters fixed, this single change reduces VoxCeleb1-H EER from 3.17 percent (SupCon) to 2.70 percent, and combining it with the H-SCL hardening loss reaches 2.60 percent, beating AAMSoftmax and H-SCL. The 18 percent relative improvement is the headline result.

Load-bearing premise

The load-bearing premise is that clustering speakers by their baseline voiceprints groups genuinely confusable speakers, so that pairs from different speakers inside one cluster are hard negatives; if the baseline embeddings do not capture speaker similarity, the clusters become arbitrary and the sampler loses its rationale.

Editorial extensions

If this is right

Existing supervised contrastive speaker-verification pipelines may be under-using hard negatives; a sampler-level change can recover those gains without retuning the loss.
The method should transfer to other encoder architectures and to applications where confusable speakers matter, such as smart-home voice assistants distinguishing family members.
Because CHNS avoids a classification layer, it scales better to training sets with very many speakers; on the internal dataset, the classification layer alone would have exceeded 23 million parameters.
CHNS and loss-based hardening (H-SCL) are complementary, since combining them yields the best results.

Reading between the lines

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

A direct stress test the paper does not run: replace the K-Means clusters with random partitions of the same size and same hard ratio; if the gains vanish, clustering is the active ingredient, and if they persist, the mechanism is batch diversity rather than hardness.
The method implies a feedback loop: better baseline embeddings produce more meaningful clusters, so periodically recomputing clusters during training could yield further gains.
The clusters are defined in the embedding space of a baseline SupCon model, which may not match the hardest negatives of the final model; measuring the overlap between within-cluster pairs and the trained model's hardest negatives would test the method's core premise.

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.

Desk Editor's Note

read the letter

Quick take: this is a clean, honest empirical paper that deserves a proper review, but the central mechanism claim needs a control experiment before I'd trust the attribution.

What's new: instead of weighting negatives in the loss (H-SCL), they cluster speaker voiceprints from a baseline SupCon model and compose batches from within-cluster speakers so that cross-speaker pairs in the batch are genuinely hard. That's a simple idea, and they execute it cleanly: code released, same model/data/training setup across all comparisons, results on three eval sets and two lightweight architectures. The gains are consistent and the relative improvements are real in magnitude (EER 3.17 to 2.70 on VoxCeleb1-H for SupCon+CHNS; 21.72 to 16.61 on CNCeleb(E)). The ablation showing CHNS works best from the start, and the negative-similarity distribution in Fig 2, support the basic story. I also appreciate that they compare against H-SCL and AAMSoftmax rather than just their own baseline.

Soft spots:

1. The mechanism is asserted more than shown. The premise in Sec 2.1 that K-means voiceprint clusters contain confusable speakers is never directly validated. Fig 2 shows CHNS batches have higher mean negative similarity, but you computed those similarities with the same kind of embeddings you clustered, so the shift is partly a restatement of the construction. A random-partition control (same batch structure, speakers assigned to pseudo-clusters arbitrarily) is the missing experiment. If CHNS beats that, the hardness interpretation holds. If not, the gain may just come from reduced speaker diversity per batch.

2. Hyperparameters K and hard_ratio are selected on VoxCeleb1-H, the same split used for the headline comparison (Table 1 then Table 2). That's selection-on-the-test-set, and the paper doesn't disclose it as such. The default CHNS row in Table 2 is the best pick from that table. They need either a held-out validation split or a statement that the chosen values are fixed a priori.

3. No error bars or multiple seeds. The differences between H-SCL and SupCon+CHNS are 0.1-0.3% EER in places; with batch size 1300 and 200 epochs those could be within run-to-run noise. They report single runs.

4. The 'dataset and model agnostic' claim outruns the evidence: one extra internal dataset and two lightweight architectures. Fine for an Interspeech paper, but soften the conclusion.

Also, they cite Go et al. [17] (hard negative sampling for SV) but never compare against it; that's a directly related method and the comparison should be there or explained.

Bottom line: if I were the editor I'd send it to review. The core empirical result is valuable and reproducible, and the missing controls are fixable. But as it stands I'd treat the 18% relative improvement as an upper bound with selection effects, not a measured property of hard-negative sampling.

Referee Report

3 major / 4 minor

Summary. The paper proposes CHNS, a batch-sampling method for supervised contrastive speaker verification. Speaker voiceprints are computed with a pre-trained SupCon model, K-means clusters group similar speakers, and the batch sampler fills a tunable fraction of each batch with speakers from one or a few clusters so that cross-speaker within-cluster pairs act as hard negatives. The contrastive loss is unchanged; only batch composition is modified. Experiments on VoxCeleb2-trained lightweight ECAPA-TDNN and Thin ResNet-34 models report consistent EER/minDCF improvements over AAMSoftmax, SupCon, and H-SCL on VoxCeleb1-H, CNCeleb(E), and an internal Bixby Eval set, with the best configuration reaching 2.60% EER on VoxCeleb1-H (H-SCL + CHNS). The code is released.

Significance. If the improvements are robust, CHNS is a valuable and simple contribution: it shows that negative-pair hardness can be controlled at the sampling level without changing the contrastive loss, and the gains replicate across three evaluation sets and two lightweight architectures. The cross-dataset results on CNCeleb(E) and Bixby Eval are particularly informative because the CHNS hyperparameters were not tuned on those sets. The open-source code and the clean experimental framework are strengths. However, the paper currently lacks a control that isolates clustering from mere partitioning, and the headline VoxCeleb1-H numbers are obtained after selecting hyperparameters on that same test set; these issues must be resolved before the central attribution claim is established.

major comments (3)

[Section 2.1, Figure 2] The central mechanism is that K-means voiceprint clusters contain genuinely confusable speakers, making within-cluster cross-speaker pairs hard negatives. The only direct evidence offered is Figure 2, which shows that CHNS batches have a higher mean negative similarity than random batches. Because the clusters are built from the same pre-trained embeddings used to compute that similarity, the distribution shift is partly a consequence of the clustering construction rather than independent evidence of hardness. Add a random-partition control: assign speakers to the same number and sizes of groups uniformly at random, use the same batch-composition rule with hard_ratio = 1, and report EER/minDCF. This is required to attribute the gains in Tables 2 and 4 to cluster-based hardness rather than to reduced speaker diversity or the specific grouped-batch structure.
[Table 1 and Table 2] The number of clusters and the hard ratio are selected by evaluating all 16 combinations on VoxCeleb1-H (Table 1), and the same dataset is then used as the first evaluation set in Table 2 to claim an 18% relative improvement. This makes the VoxCeleb1-H comparison a post-selection result. The selection should be performed on a separate validation split or with nested evaluation, and the paper should clearly state which reported numbers are post-selection. The CNCeleb(E) and Bixby Eval results are not affected by this issue, but the headline VoxCeleb1-H number is.
[Tables 1-4] No measure of run-to-run variability is reported. Since the differences between some configurations are small (e.g., Table 4, SupCon + CHNS at 2.94% EER versus AAMSoftmax at 3.17%, and Table 1 adjacent cells differing by about 0.1% EER), a single run cannot establish the significance of the rankings. Report mean and standard deviation over at least three training runs for the main comparisons, or provide a paired significance test.

minor comments (4)

[Section 3.2] The text contains minor typographical issues: 'Detection Cost Fuction' should be 'Detection Cost Function', and 'V oxCeleb'/'V oxCeleb2' appear with stray spaces in several places.
[Section 4] The notation 'HSCL + CHNS' and 'H-SCL + CHNS' is used interchangeably; please choose one consistent form.
[Figure 2] The caption should state whether the similarity distribution is computed with the pre-trained baseline embeddings used for clustering or with the model being trained; otherwise the reader cannot directly assess the potential circularity in the evidence.
[Table 3] The term 'curriculum learning' is imprecise for experiments that introduce CHNS at fixed epochs rather than gradually increasing difficulty; 'delayed CHNS introduction' would be more accurate.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the proposed method's claims are empirical and evaluated on held-out data.

full rationale

The paper proposes a batch-sampling algorithm (CHNS) that uses K-means clusters of speaker voiceprints from a separately trained SupCon baseline to compose training batches with within-cluster negative pairs. The central claim, that changing only batch composition improves speaker verification, is tested on held-out evaluation sets (VoxCeleb1-H, CNCeleb(E), and an in-house Bixby Eval) against independent baselines (AAMSoftmax, SupCon, H-SCL). No result is derived from the method's own output: the clustering input comes from an independently trained baseline model, and the trained model is not used to define the evaluation metric. The assumption that within-cluster cross-speaker pairs are hard negatives is an untested premise rather than a circular step, and the VoxCeleb1-H hyperparameter selection creates selection bias but not circular reasoning. There are no self-citations carrying load-bearing arguments, no fitted parameters renamed as predictions, and no uniqueness claims imported from the authors' prior work.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The ledger is small: two tuned hyperparameters (cluster count and hard ratio) are the main free parameters, and both were selected on the VoxCeleb1-H evaluation set, which introduces selection bias. The core domain assumption is that voiceprint clustering identifies genuinely confusable speakers. No new physical or algorithmic entities are invented.

free parameters (4)

number of clusters K = 50
Chosen as best on the VoxCeleb1-H evaluation set (Table 1). Controls average cluster size and thus the hardness of within-cluster negative pairs.
hard_ratio = 1.0
Chosen as best on VoxCeleb1-H (Table 1). Determines the fraction of each batch drawn from clusters; 1.0 means all batch speakers come from clusters.
H-SCL hardening exponent beta = 0.1
Set experimentally in Section 3.3 for the loss-based hard negative sampling baseline and for the combined H-SCL + CHNS variant.
utterances per speaker for voiceprint = 10
A design choice in Section 2.1; no sensitivity analysis is provided for this count.

assumptions (4)

domain assumption K-means clustering on L2-normalized voiceprint centroids groups speakers such that within-cluster pairs are hard negatives.
Section 2.1 states that pairs from different within-cluster speakers can be considered hard negatives. This is the core mechanistic premise; if clusters are not perceptually meaningful, the method loses its foundation.
domain assumption A centroid of 10 random utterances per speaker is a stable and representative voiceprint.
Section 2.1 computes voiceprints from a small sample of utterances without quantifying variance or stability across utterances.
domain assumption The baseline SupCon model trained with random sampling provides embeddings suitable for clustering similar speakers.
Section 2.1 uses a pretrained supervised contrastive model to obtain voiceprints. The success of CHNS depends on this initialization capturing speaker similarity.
standard math Equation (1) with beta=0 is equivalent to the SupCon loss and, with one positive pair per speaker, to the N-pair loss.
Stated in Section 2.2, used for the baseline comparison.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Clustering-based hard negative sampling for supervised contrastive speaker verification." pith.science (2026). https://pith.science/paper/ATH4FS7I

@misc{pith2026250717540,
  author       = {Pith},
  title        = {Pith review of: Clustering-based hard negative sampling for supervised contrastive speaker verification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ATH4FS7I}},
  note         = {Machine review of arXiv:2507.17540}
}

read the original abstract

In speaker verification, contrastive learning is gaining popularity as an alternative to the traditionally used classification-based approaches. Contrastive methods can benefit from an effective use of hard negative pairs, which are different-class samples particularly challenging for a verification model due to their similarity. In this paper, we propose CHNS - a clustering-based hard negative sampling method, dedicated for supervised contrastive speaker representation learning. Our approach clusters embeddings of similar speakers, and adjusts batch composition to obtain an optimal ratio of hard and easy negatives during contrastive loss calculation. Experimental evaluation shows that CHNS outperforms a baseline supervised contrastive approach with and without loss-based hard negative sampling, as well as a state-of-the-art classification-based approach to speaker verification by as much as 18 % relative EER and minDCF on the VoxCeleb dataset using two lightweight model architectures.

Figures

Figures reproduced from arXiv: 2507.17540 by the authors.

**Figure 1.** Proposed within-batch similarity matrix with hard ratio set to 0.8. C1, C2 - cluster sizes, BS - batch size, N - negative, HN - hard negative, P - positive, (xn, xn’) - same speaker utterance pair. speakers belonging to specific clusters. We start with an empty batch, and sample the first speaker cluster randomly. If a sampled cluster contains less than hard ratio ∗ batch size speakers, another random cluster is … view at source ↗

**Figure 2.** Average distribution of negative pair similarities in the training batch depending on the batch sampling method. max), a supervised contrastive model with random sampling (SupCon), and a supervised contrastive model with loss-based hard negative sampling (H-SCL) [19]. To push the limits of hard negative sampling, as a last solution we also combine the H-SCL method with CHNS (denoted as H-SCL + CHNS). Table 2 presen… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references · 28 canonical work pages

[17]

The idlab voxceleb speaker recognition challenge 2020 system description,

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

arXiv 2020
[1]

Clustering-based hard negative sampling for supervised contrastive speaker verification

Introduction Speaker verification (SV) is a task that aims to verify the iden- tity of a speaker based on voice characteristics. In recent years, researchers have been improving the performance of SV sys- tems using various deep learning approaches with great success [1, 2, 3, 4, 5]. The standard method involves learning a fixed- size speaker embedding, w...

work page Pith review arXiv 2025
[2]

We use a contrastive loss function during the training pro- cess, which calculates the relationship between speaker repre- sentations on a within-batch basis

Proposed Method Our proposed solution uses supervised contrastive learning to train a model that produces utterance-level speaker representa- tions. We use a contrastive loss function during the training pro- cess, which calculates the relationship between speaker repre- sentations on a within-batch basis. This means that batch com- position has a signifi...

work page 1903
[3]

For a fair comparison, in all experiments, we use the exact same model, data and training parameters

Experimental setup and datasets We conduct a comprehensive experimental evaluation to find the best parameters for the proposed CHNS algorithm, and compare our solution to existing training methods: a supervised contrastive approach with random batch sampling (SupCon), a supervised contrastive approach with loss-based hard nega- tive sampling (H-SCL), and...

work page
[4]

The two variables in the algorithm are the number of clusters (which also correlates with the average cluster size) and hard ratio

Experiments and results In the first experiment we establish the best parameters for CHNS. The two variables in the algorithm are the number of clusters (which also correlates with the average cluster size) and hard ratio. We test 4 cluster numbers of 10, 20, 50 and 100, and 4 hard ratio values: 0.2, 0.5, 0.8 and 1. We compare results on the V oxCeleb1-H ...

work page 1912
[5]

Conclusion In this paper, we proposed CHNS - a clustering-based hard neg- ative sampling approach for supervised contrastive speaker ver- ification. Through a series of experiments using lightweight models suitable for use on edge devices, we demonstrated that our method outperforms standard supervised contrastive learn- ing with and without loss-based ha...

work page
[6]

Deep Neural Network Embeddings for Text-Independent Speaker Verification,

D. Snyder, D. Garcia-Romero, D. Povey, and S. Khudan- pur, “Deep Neural Network Embeddings for Text-Independent Speaker Verification,” in Proc. Interspeech 2017, 2017, pp. 999– 1003

work page 2017
[7]

Densely Connected Time Delay Neu- ral Network for Speaker Verification,

Y .-Q. Yu and W.-J. Li, “Densely Connected Time Delay Neu- ral Network for Speaker Verification,” in Proc. Interspeech 2020, 2020, pp. 921–925

work page 2020

Show all 33 references

[8]

X-vectors: Robust dnn embeddings for speaker recognition,

D. Snyder, D. Garcia-Romero, G. Sell, D. Povey, and S. Khudan- pur, “X-vectors: Robust dnn embeddings for speaker recognition,” in 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2018, pp. 5329–5333

2018
[9]

ECAPA-TDNN: Emphasized channel attention, propaga- tion and aggregation in TDNN based speaker verifica- tion,

B. Desplanques, J. Thienpondt, and K. Demuynck, “ECAPA-TDNN: Emphasized channel attention, propaga- tion and aggregation in TDNN based speaker verifica- tion,” in Interspeech 2020 , Oct. 2020. [Online]. Available: http://dx.doi.org/10.21437/Interspeech.2020-2650

2020 doi
[10]

Wavlm: Large-scale self-supervised pre-training for full stack speech processing,

S. Chen, C. Wang, Z. Chen, Y . Wu, S. Liu, Z. Chen, J. Li, N. Kanda, T. Yoshioka, X. Xiao, J. Wu, L. Zhou, S. Ren, Y . Qian, Y . Qian, M. Zeng, and F. Wei, “Wavlm: Large-scale self-supervised pre-training for full stack speech processing,” IEEE Journal of Selected Topics in Si...

2021
[11]

Arcface: Additive angular margin loss for deep face recognition,

J. Deng, J. Guo, N. Xue, and S. Zafeiriou, “Arcface: Additive angular margin loss for deep face recognition,” in2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019, pp. 4685–4694

2019
[12]

Discriminative speaker representation via contrastive learning with class-aware atten- tion in angular space,

Z. Li, M.-W. Mak, and H. M.-L. Meng, “Discriminative speaker representation via contrastive learning with class-aware atten- tion in angular space,” in ICASSP 2023 - 2023 IEEE Interna- tional Conference on Acoustics, Speech and Signal Processing (ICASSP), 2023, pp. 1–5

2023
[13]

Contrastive learning for improving end-to-end speaker verification,

Y . Tang, J. Wang, X. Qu, and J. Xiao, “Contrastive learning for improving end-to-end speaker verification,” in2021 International Joint Conference on Neural Networks (IJCNN), 2021, pp. 1–7

2021
[14]

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,” inProc. Interspeech 2020, 2020, pp. 2977– 2981

2020
[15]

Contrastive self-supervised learning for text-independent speaker verification,

H. Zhang, Y . Zou, and H. Wang, “Contrastive self-supervised learning for text-independent speaker verification,” in ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2021, pp. 6713–6717

2021
[16]

Label-Efficient Self-Supervised Speaker Verification With Information Maximization and Con- trastive Learning,

T. Lepage and R. Dehak, “Label-Efficient Self-Supervised Speaker Verification With Information Maximization and Con- trastive Learning,” in Proc. Interspeech 2022 , 2022, pp. 4018– 4022

2022
[18]

Self-supervised speaker recognition with loss-gated learning,

R. Tao, K. Aik Lee, R. Kumar Das, V . Hautam ¨aki, and H. Li, “Self-supervised speaker recognition with loss-gated learning,” in ICASSP 2022 - 2022 IEEE International Conference on Acous- tics, Speech and Signal Processing (ICASSP) , 2022, pp. 6142– 6146

2022
[19]

Experimenting with Additive Margins for Contrastive Self-Supervised Speaker Verification,

T. Lepage and R. Dehak, “Experimenting with Additive Margins for Contrastive Self-Supervised Speaker Verification,” in Proc. INTERSPEECH 2023, 2023, pp. 4708–4712

2023
[20]

A simple framework for contrastive learning of visual representations,

T. Chen, S. Kornblith, M. Norouzi, and G. Hinton, “A simple framework for contrastive learning of visual representations,” in Proceedings of the 37th International Conference on Machine Learning, ser. ICML’20. JMLR.org, 2020

2020
[21]

Supervised contrastive learning,

P. Khosla, P. Teterwak, C. Wang, A. Sarna, Y . Tian, P. Isola, A. Maschinot, C. Liu, and D. Krishnan, “Supervised contrastive learning,” in Proceedings of the 34th International Conference on Neural Information Processing Systems, ser. NIPS ’20. Red Hook, NY , USA: Curran Asso...

2020
[22]

Contrastive speaker representation learning with hard negative sampling for speaker recognition,

C. Go, Y . H. Lee, T. Kim, N. I. Park, and C. Chun, “Contrastive speaker representation learning with hard negative sampling for speaker recognition,” Sensors, vol. 24, no. 19, 2024

2024
[23]

Contrastive learning with hard negative samples,

J. D. Robinson, C.-Y . Chuang, S. Sra, and S. Jegelka, “Contrastive learning with hard negative samples,” inInternational Conference on Learning Representations , 2021. [Online]. Available: https: //openreview.net/forum?id=CR1XOQ0UTh-

2021
[24]

Supervised con- trastive learning with hard negative samples,

R. Jiang, T. Nguyen, P. Ishwar, and S. Aeron, “Supervised con- trastive learning with hard negative samples,” in 2024 Interna- tional Joint Conference on Neural Networks (IJCNN) , 2024, pp. 1–8

2024
[25]

SimCSE: Simple contrastive learn- ing of sentence embeddings,

T. Gao, X. Yao, and D. Chen, “SimCSE: Simple contrastive learn- ing of sentence embeddings,” in Empirical Methods in Natural Language Processing (EMNLP), 2021

2021
[26]

V oice biometrical match of twin and non-twin siblings,

P. Gomez and E. San Segundo, “V oice biometrical match of twin and non-twin siblings,” in Proceedings of the 8th International Workshop Models and analysis of vocal emissions for biomedical applications, 06 2013

2013
[27]

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
[28]

Improving dino-based self-supervised speaker verification with progressive cluster- aware training,

B. Han, W. Huang, Z. Chen, and Y . Qian, “Improving dino-based self-supervised speaker verification with progressive cluster- aware training,” in 2023 IEEE International Conference on Acoustics, Speech, and Signal Processing Workshops (ICASSPW), 2023, pp. 1–5

2023
[29]

Cluster-guided unsupervised domain adaptation for deep speaker embedding,

H. Mao, F. Hong, and M.-w. Mak, “Cluster-guided unsupervised domain adaptation for deep speaker embedding,” IEEE Signal Processing Letters, vol. 30, pp. 643–647, 2023

2023
[30]

A triangle inequality for cosine similarity,

E. Schubert, “A triangle inequality for cosine similarity,” in Sim- ilarity Search and Applications, N. Reyes, R. Connor, N. Kriege, D. Kazempour, I. Bartolini, E. Schubert, and J.-J. Chen, Eds. Cham: Springer International Publishing, 2021, pp. 32–44

2021
[31]

Improved deep metric learning with multi-class n-pair loss objective,

K. Sohn, “Improved deep metric learning with multi-class n-pair loss objective,” in Advances in Neural Information Processing Systems, D. Lee, M. Sugiyama, U. Luxburg, I. Guyon, and R. Garnett, Eds., vol. 29. Curran Associates, Inc., 2016. [Online]. Available: https://proceedi...

2016
[32]

V oxCeleb2: Deep Speaker Recognition,

J. S. Chung, A. Nagrani, and A. Zisserman, “V oxCeleb2: Deep Speaker Recognition,” inProc. Interspeech 2018, 2018, pp. 1086– 1090

2018
[33]

Cn-celeb: a challenging chinese speaker recognition dataset,

Y . Fan, J. Kang, L. Li, K. Li, H. Chen, S. Cheng, P. Zhang, Z. Zhou, Y . Cai, and D. Wang, “Cn-celeb: a challenging chinese speaker recognition dataset,” inIEEE ICASSP 2020. IEEE, 2020, pp. 7604–7608

2020

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [17]

The idlab voxceleb speaker recognition challenge 2020 system description,

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

arXiv 2020

[2] [1]

Clustering-based hard negative sampling for supervised contrastive speaker verification

Introduction Speaker verification (SV) is a task that aims to verify the iden- tity of a speaker based on voice characteristics. In recent years, researchers have been improving the performance of SV sys- tems using various deep learning approaches with great success [1, 2, 3, 4, 5]. The standard method involves learning a fixed- size speaker embedding, w...

work page Pith review arXiv 2025

[3] [2]

We use a contrastive loss function during the training pro- cess, which calculates the relationship between speaker repre- sentations on a within-batch basis

Proposed Method Our proposed solution uses supervised contrastive learning to train a model that produces utterance-level speaker representa- tions. We use a contrastive loss function during the training pro- cess, which calculates the relationship between speaker repre- sentations on a within-batch basis. This means that batch com- position has a signifi...

work page 1903

[4] [3]

For a fair comparison, in all experiments, we use the exact same model, data and training parameters

Experimental setup and datasets We conduct a comprehensive experimental evaluation to find the best parameters for the proposed CHNS algorithm, and compare our solution to existing training methods: a supervised contrastive approach with random batch sampling (SupCon), a supervised contrastive approach with loss-based hard nega- tive sampling (H-SCL), and...

work page

[5] [4]

The two variables in the algorithm are the number of clusters (which also correlates with the average cluster size) and hard ratio

Experiments and results In the first experiment we establish the best parameters for CHNS. The two variables in the algorithm are the number of clusters (which also correlates with the average cluster size) and hard ratio. We test 4 cluster numbers of 10, 20, 50 and 100, and 4 hard ratio values: 0.2, 0.5, 0.8 and 1. We compare results on the V oxCeleb1-H ...

work page 1912

[6] [5]

Conclusion In this paper, we proposed CHNS - a clustering-based hard neg- ative sampling approach for supervised contrastive speaker ver- ification. Through a series of experiments using lightweight models suitable for use on edge devices, we demonstrated that our method outperforms standard supervised contrastive learn- ing with and without loss-based ha...

work page

[7] [6]

Deep Neural Network Embeddings for Text-Independent Speaker Verification,

D. Snyder, D. Garcia-Romero, D. Povey, and S. Khudan- pur, “Deep Neural Network Embeddings for Text-Independent Speaker Verification,” in Proc. Interspeech 2017, 2017, pp. 999– 1003

work page 2017

[8] [7]

Densely Connected Time Delay Neu- ral Network for Speaker Verification,

Y .-Q. Yu and W.-J. Li, “Densely Connected Time Delay Neu- ral Network for Speaker Verification,” in Proc. Interspeech 2020, 2020, pp. 921–925

work page 2020