Pith. sign in

REVIEW 4 major objections 7 minor 35 references

Towards Early Prediction of Self-Supervised Speech Model Performance

T0 review · 4 major / 7 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Unsupervised measures of rank and cluster quality of SSL speech embeddings predict final downstream performance better than the pre-training loss does.

desk verdict The central empirical finding is real, but the headline correlation is inflated by choosing the layer after seeing the labels; still worth a serious referee. read the letter →

arxiv 2501.05966 v2 pith:IXS4NSLL submitted 2025-01-10 cs.SD cs.CLcs.LGeess.AS

Ryan Whetten , Lucas Maison , Titouan Parcollet , Marco Dinarelli , Yannick Estève This is my paper · ORCID
classification cs.SDcs.CLcs.LGeess.AS
keywords self-supervisedlearningspeechrepresentationeffectiverankmodelevaluationclusteringBEST-RQautomaticrecognitionearlyprediction
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 establish that the quality of a self-supervised speech model can be judged without labeled data or fine-tuning, by looking at the geometry of its embeddings on one hour of unlabeled audio. It trains 30 BEST-RQ models with different batch sizes and mask rates and shows that the global effective rank of the embeddings at 50,000 pre-training steps correlates with the final in-domain ASR word error rate at about $-0.9$, while the pre-training loss correlates at only about $-0.3$. The same holds, more weakly, for clustering-based measures and for speaker verification. If the claim holds, model selection during pre-training can be done cheaply on CPUs, potentially saving thousands of GPU hours and removing the need for labeled validation data.

What carries the argument

The Global Effective Rank (GER): concatenate all frame-level embeddings from a layer across about one hour of audio, then take the effective rank $E\text{-Rank}(A)=\exp(-\sum_i p_i\log p_i)$, where $p_i$ are normalized singular values. This is a label-free proxy for how much information the representation carries. The argument also uses MiniBatchKMeans clustering to compute inertia (within-cluster sum of squares) and the Davies-Bouldin index, which measure dispersion and cluster separation; the paper interprets low inertia as good organization and notes the DB index can behave counter-intuitively because overlapping clusters can still be useful.

What would settle it

Fix a layer before seeing any downstream labels (for example, always use layer 6), compute GER at 50k steps on a held-out set of BEST-RQ runs, and measure its rank correlation with final WER. If the correlation is no better than the pre-training loss, the unsupervised early-prediction claim fails for the layer-agnostic version.

Watch

Extended reading notes

Core claim

The central claim is that the global effective rank (the exponential of the entropy of the normalized singular values of the matrix formed by concatenating all frame-level embeddings) is a better early indicator of future downstream performance than the pre-training loss. For in-domain LibriSpeech ASR, the GER computed from layer 12 embeddings at 50k steps correlates with the final WER at 200k steps with a Pearson coefficient of about -0.9, whereas the pre-training loss gives about -0.3. Cluster-quality measures (inertia and the Davies-Bouldin index) also beat the loss but are weaker than rank measures. For out-of-domain ASR and speaker verification, the correlations are weaker, particularly once all models converge to similar performance.

Load-bearing premise

The method's reported correlations rely on measuring embeddings at the layer that was later found to correlate best with the downstream labels (layer 12 for ASR, layer 8 for SV); no unsupervised rule is given for choosing that layer. If the best layer must be identified with labeled data, a major part of the labeled-data cost the method claims to avoid is still present.

Editorial extensions

If this is right

  • During pre-training, a practitioner can rank candidate runs by GER at 50k steps and keep only the top few, avoiding full training of all candidates; the paper estimates savings of about 10,800 GPU hours on its 30-model setup.
  • The pre-training loss should not be used as a quality gate for SSL speech models, since mask rate and batch size change task difficulty and move the loss independently of downstream quality.
  • For in-domain ASR, even a 50k-step checkpoint with only 5 downstream epochs gives a WER that correlates at about 0.95-0.97 with the final 200k-step, 20-epoch result, so labeled early stopping is also possible when labels exist.
  • The unsupervised measures are most informative when candidate models actually differ in performance; when models converge to similar WER/error rates, correlations weaken.

Reading between the lines

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

  • Going beyond the paper: the layer-selection problem could be attacked by aggregating GER across layers or using a fixed early layer, which would make the pipeline fully unsupervised; the paper does not test this.
  • Going beyond the paper: if effective rank tracks information content, GER could also be monitored along the training curve of a single model to decide when to stop pre-training, not just to compare checkpoints across models.
  • Going beyond the paper: the same cost-geometry idea may transfer to other SSL speech architectures and to non-ASR tasks, but the paper only tests BEST-RQ with ASR and SV, so transfer remains an open empirical question.
  • Going beyond the paper: because correlations are computed over 30 models drawn from a deliberately varied hyperparameter grid, the reported magnitudes could shrink on a more homogeneous set of runs; a useful test is rerunning on a hold-out design with confidence intervals.
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

4 major / 7 minor

Summary. The paper proposes unsupervised, label-free measures—cluster quality (inertia, Davies-Bouldin) and embedding rank (RankMe-t and a newly proposed Global Effective Rank, GER)—computed from one hour of unlabeled audio, as early indicators of downstream performance for self-supervised speech models. The authors train 30 BEST-RQ models varying batch size and mask proportion, and correlate the unsupervised measures with ASR (in-domain LibriSpeech, out-of-domain CommonVoice) and speaker verification (VoxCeleb) performance. They report that GER at layer 12 for ASR correlates around -0.9 with final WER, versus about -0.3 for the pre-training loss, and argue that such measures could save thousands of GPU hours by enabling early selection of promising checkpoints.

Significance. If the central claim is established under a genuinely unsupervised protocol, the work would provide a compute- and data-efficient early-stopping or model-selection signal for SSL speech pre-training, with substantial practical value. The paper has clear strengths: the code is publicly released, the comparison against the pre-training loss baseline is appropriate, the proposed GER is a natural extension of RankMe-t, and the evaluation covers multiple downstream tasks. However, the post hoc layer selection and the lack of uncertainty quantification mean that the current evidence does not yet support the advertised 'unsupervised' claim; the reported headline correlations are in-sample maxima over layers rather than the output of a fixed label-free protocol.

major comments (4)
  1. [Section 4.4] The layer selection is post hoc and undermines the unsupervised claim. The text states that for ASR layer 12 and for SV layer 8 were chosen because the measures computed on those layers correlated most with downstream performance. Consequently, the headline correlations in Tables 1 and 2 (e.g., GER -0.905 in Table 2) are the maximum over layers of an in-sample selection process, not the result of a reproducible unsupervised protocol. To support the claimed label-free early prediction, the paper should either report the correlations for all layers (e.g., a curve or heatmap) so that the sensitivity to layer choice is visible, or specify and validate a fixed rule for choosing the layer without downstream labels (for example, always using the last layer or a layer chosen by an unsupervised criterion). Without this, the practical method remains underspecified and the reported -0.9 correlations are an upper bound from post hoc selection.
  2. [Table 2 and Section 4.1] The reported Pearson correlations lack uncertainty quantification, which is important given only n=30 models and a hyperparameter grid varying only two axes (batch size and mask proportion). The difference between GER (-0.905) and pre-training loss (-0.324) could be driven by a small number of points visible in Figure 1, and the optimal checkpoint (50k vs 200k) and the optimal metric are also selected by looking at the results. Provide confidence intervals (e.g., bootstrap or Fisher z-transformation) for the correlations, and report the spread of correlations across layers, checkpoints, and metrics to demonstrate that the conclusion is robust to these choices rather than being an artifact of selection.
  3. [Section 2.2] The theoretical argument is an illustrative extreme case, not a general derivation. The example with m_a near 0 and m_b larger shows that a higher loss can coincide with better downstream performance in a specific regime, but it does not establish that the pre-training loss is a poor indicator across the range of hyperparameters studied in the experiments. The section should be framed as motivation or intuition, not as a theoretical basis for discarding the loss; the empirical comparison in Section 4 is the actual evidence and should carry the weight of the claim.
  4. [Section 4.3 and Table 1] The out-of-domain ASR and speaker verification results are weak and inconsistent in sign across checkpoints (e.g., inertia -0.000 at 50k vs 0.367 at 200k for VoxCeleb; GER 0.163 at 50k vs 0.584 at 200k). The paper attributes this to similar downstream performance among models, but it does not quantify the performance spread. Report the range or standard deviation of the downstream metrics for each task, and temper the abstract's general claim that these measures predict 'SSL speech model performance' when the evidence is strong only for in-domain ASR.
minor comments (7)
  1. [Section 3.1.1] There are typos in this section: 'custers' should be 'clusters' and 'calulated' should be 'calculated'.
  2. [Section 2.3] 'looses' should be 'loses', and 'we hypothesis' should be 'we hypothesize'.
  3. [Section 4.2] Several typos appear: 'sugests' should be 'suggests', 'guage' should be 'gauge', and 'finial' should be 'final'.
  4. [Equation (4)] The notation for the concatenation is malformed: the displayed expression 'C = n i=0 Si' is not clear. Please define C as the concatenation of all embedding sequences with standard notation.
  5. [Table 1] The row label 'Global Rank (GER)' should be 'Global Effective Rank (GER)' for consistency with Section 3.1.2.
  6. [Section 4.1] The preliminary experiment shows that the global effective rank plateaus with one hour of audio, but the same one-hour budget is used for all measures; please state whether the clustering measures also plateau at that amount or at least acknowledge this is assumed.
  7. [Discussion] The final paragraph contains a duplicated and incomplete sentence: 'We believe that future work could involve exploring these methods further and developing.' This should be removed or completed.

Circularity Check

1 steps flagged · score 3.0 of 10

Headline correlations are partly in-sample because the layer (12 for ASR, 8 for SV) was selected by inspecting which layer correlated best with downstream labels; the core metrics themselves are label-free, so this is a selection/peeking issue rather than a derivation-level circularity.

  1. other [Section 4.4, Additional observations; used for Table 1 and Table 2]
    "We report the correlations from measures calculated on the outputs of layer 12 for ASR and layer 8 for SV in both Table 1 and 2."

    The paper's headline result is that unsupervised measures computed at 50k steps, especially GER, correlate with final WER at about -0.9 (Table 2). However, the layer used in the reported tables was chosen after the fact because it had the strongest correlation with downstream performance: the preceding paragraph says 'we found that the measures computed on the embeddings extracted from the first and last layers correlated the most with downstream performance' and 'For SV, we found that the middle layers tended to have stronger correlations.' No unsupervised rule for selecting the layer is given.

full rationale

The paper's unsupervised measures (inertia, Davies-Bouldin index, RankMe-t, GER) are computed without downstream labels, and the correlations across 30 trained models are genuine empirical findings, not definitions. There is no self-definitional step: GER is defined directly from embedding concatenation, and WER comes from separate downstream recipes. The paper does not rely on a load-bearing self-citation chain: its own prior work is cited only for efficiency, open implementations, or architectural choices, and the rank-based methodology originates from external work (Garrido et al., Aldeneh et al.). The main circularity-adjacent issue is the post hoc selection of the reporting layer in Section 4.4: layer 12 (ASR) and layer 8 (SV) were chosen because they correlated most with downstream performance, and then those same correlations are reported as evidence. This makes the headline -0.9 correlations partly in-sample and means the method as advertised ('unsupervised early prediction') is not fully specified without a label-free layer-selection rule. Because the metric computation itself is label-free and the correlation is not forced by construction, this warrants a moderate score rather than a high one.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The central claim rests on the empirical hypothesis that intrinsic embedding statistics predict downstream performance. It is supported only by correlational evidence from 30 BEST-RQ models, with post hoc layer selection and no significance testing.

free parameters (3)
  • Number of KMeans clusters (k) = 1024
    Fixed without sensitivity analysis; WCSS and DB values depend on k.
  • Embedding layer index = 12 (ASR), 8 (SV)
    Selected post hoc as the layer with strongest correlation to downstream labels; this inflates the reported correlations.
  • Amount of unlabeled audio = 1 hour
    Chosen from preliminary plateau experiments for GER; no error bars or dataset details are shown.
assumptions (3)
  • domain assumption Effective rank and clustering metrics on frozen embeddings are meaningful indicators of representation quality
    This is the central hypothesis; the paper provides only correlational evidence, no mechanistic proof.
  • domain assumption The 30 BEST-RQ models with varied batch size and mask percentage are representative of SSL speech training conditions
    All models share the same architecture and pre-training data; generalization to other SSL methods is untested.
  • domain assumption The illustrative example in Section 2.2 (low mask leads to mimicking the quantizer, high mask leads to better statistical learning) justifies why loss is not a reliable indicator
    The argument covers an extreme constructed case (m_a approximately 0) and is not a general proof.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Early Prediction of Self-Supervised Speech Model Performance." pith.science (2026). https://pith.science/paper/IXS4NSLL

@misc{pith2026250105966,
  author       = {Pith},
  title        = {Pith review of: Towards Early Prediction of Self-Supervised Speech Model Performance},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IXS4NSLL}},
  note         = {Machine review of arXiv:2501.05966}
}
read the original abstract

In Self-Supervised Learning (SSL), pre-training and evaluation are resource intensive. In the speech domain, current indicators of the quality of SSL models during pre-training, such as the loss, do not correlate well with downstream performance. Consequently, it is often difficult to gauge the final downstream performance in a cost efficient manner during pre-training. In this work, we propose unsupervised efficient methods that give insights into the quality of the pre-training of SSL speech models, namely, measuring the cluster quality and rank of the embeddings of the SSL model. Results show that measures of cluster quality and rank correlate better with downstream performance than the pre-training loss with only one hour of unlabeled audio, reducing the need for GPU hours and labeled data in SSL model evaluation.

Figures

Figures reproduced from arXiv: 2501.05966 by the authors.

Figure 1
Figure 1. Correlations between 4 selected measures (calculated on embeddings at 50k pre-training steps) and the future WER on the LibriSpeech dev-clean set (at 200k pre-training steps and 20 epochs fine-tuning). Each point represents a model with different hyper-parameters (batch size and mask percentage). Fig. 1a, show that some of the models with the lowest WER (y-axis) have a relatively high pre-training loss (x-axis) resu… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 19 canonical work pages

  1. [1]

    Introduction Self-Supervised Learning (SSL) is the process of training a model where the targets are derived directly from the input data itself, decreasing the need of human-labeled data. SSL mod- els are usually trained in two stages:pre-training, where the model is trained using unlabelled data, and subsequently,fine- tuning, where the model is trained...

    work page

  2. [2]

    Towards Early Prediction of Self-Supervised Speech Model Performance

    Background In this section, we give an overview of BEST-RQ, explain how the pre-training loss is calculated, and show theoretically why it is not a good indicator of downstream performance. This is supported empirically in Section 4. 2.1. Architecture of BEST-RQ We use BEST-RQ due to its efficiency, performance, and open source implementation in the Speec...

    work page Pith review arXiv 2025

  3. [3]

    Methods In this section, we introduce the unsupervised measures used in our experiments and then describe the experimental settings, including the models trained and the downstream tasks. 3.1. Unsupervised measures for SSL model evaluation We hypothesize that the dispersion (or lack thereof) of the out- put embeddings in the latent space is correlated wit...

    work page

  4. [4]

    Then we present results for each task

    Results We first report on preliminary experiments that informed our choices on the amount of audio to use for the unsupervised eval- uation methods and the number of epochs for the downstream task. Then we present results for each task. 4.1. Preliminary experiments For generating the output embeddings, we randomly sample about 1 hour of audio from each d...

    work page

  5. [5]

    To illustrate the potential compute savings, suppose one pre- trains the 30 models from this work to 400k steps

    Discussion We propose using clustering and rank measures as efficient, unsupervised methods for evaluating the performance of SSL speech models and show that these measures are better indi- cators of downstream performance than the pre-training loss. To illustrate the potential compute savings, suppose one pre- trains the 30 models from this work to 400k ...

    work page

  6. [6]

    Acknowledgements This work received funding from the French ANR E-SSL project N°ANR-22-CE23-0013 and used HPC resources from GENCI–IDRIS projects AD011014732 and A0131013821

    work page

  7. [7]

    Self-supervised speech representation learning: A review,

    A. Mohamed, H.-y. Lee, L. Borgholt, J. D. Havtorn, J. Edin, C. Igel, K. Kirchhoff, S.-W. Li, K. Livescu, L. Maaløeet al., “Self-supervised speech representation learning: A review,”IEEE Journal of Selected Topics in Signal Processing, 2022

    2022

  8. [8]

    Superb: Speech pro- cessing universal performance benchmark,

    S. wen Yang, P.-H. Chi, Y .-S. Chuang, C.-I. J. Lai, K. Lakhotia, Y . Y . Lin, A. T. Liu, J. Shi, X. Chang, G.-T. Lin, T.-H. Huang, W.-C. Tseng, K. tik Lee, D.-R. Liu, Z. Huang, S. Dong, S.-W. Li, S. Watanabe, A. Mohamed, and H. yi Lee, “Superb: Speech pro- cessing universal performance benchmark,” inInterspeech 2021, 2021, pp. 1194–1198

    2021

Show all 35 references
  1. [9]

    Reducing barriers to self-supervised learning: Hubert pre- training with academic compute,

    W. Chen, X. Chang, Y . Peng, Z. Ni, S. Maiti, and S. Watan- abe, “Reducing barriers to self-supervised learning: Hubert pre- training with academic compute,” inInterspeech 2023, 2023, pp. 4404–4408

    2023

  2. [10]

    Xls-r: Self-supervised cross-lingual speech representation learning at scale,

    A. Babu, C. Wang, A. Tjandra, K. Lakhotia, Q. Xu, N. Goyal, K. Singh, P. von Platen, Y . Saraf, J. Pino, A. Baevski, A. Con- neau, and M. Auli, “Xls-r: Self-supervised cross-lingual speech representation learning at scale,” inInterspeech 2022, 2022, pp. 2278–2282

    2022

  3. [11]

    Google usm: Scaling auto- matic speech recognition beyond 100 languages,

    Y . Zhang, W. Han, J. Qin, Y . Wang, A. Bapna, Z. Chen, N. Chen, B. Li, V . Axelrod, G. Wanget al., “Google usm: Scaling auto- matic speech recognition beyond 100 languages,”arXiv preprint arXiv:2303.01037, 2023

    2023 arXiv

  4. [12]

    Efficient self- supervised learning with contextualized target representations for vision, speech and language,

    A. Baevski, A. Babu, W.-N. Hsu, and M. Auli, “Efficient self- supervised learning with contextualized target representations for vision, speech and language,” inInternational Conference on Ma- chine Learning. PMLR, 2023, pp. 1416–1429

    2023

  5. [13]

    On the (in)efficiency of acoustic feature extrac- tors for self-supervised speech representation learning,

    T. Parcollet, S. Zhang, R. van Dalen, A. G. C. P. Ramos, and S. Bhattacharya, “On the (in)efficiency of acoustic feature extrac- tors for self-supervised speech representation learning,” inInter- speech 2023, 2023, pp. 581–585

    2023

  6. [14]

    An analysis of linear complexity attention substitues with best- rq,

    R. Whetten, T. Parcollet, A. Moumen, M. Dinarelli, and Y . Est`eve, “An analysis of linear complexity attention substitues with best- rq,” inSpoken Language Technoloby Workshop, 2024

    2024

  7. [15]

    Speech self-supervised representation benchmarking: Are we doing it right?

    S. Zaiem, Y . Kemiche, T. Parcollet, S. Essid, and M. Ravanelli, “Speech self-supervised representation benchmarking: Are we doing it right?” inInterspeech 2023, 2023, pp. 2873–2877

    2023

  8. [16]

    RankMe: Assessing the downstream performance of pretrained self-supervised representations by their rank,

    Q. Garrido, R. Balestriero, L. Najman, and Y . Lecun, “RankMe: Assessing the downstream performance of pretrained self-supervised representations by their rank,” inProceedings of the 40th International Conference on Machine Learning, ser. Proceedings of Machine Learning Resear...

    2023

  9. [17]

    The effective rank: A measure of ef- fective dimensionality,

    O. Roy and M. Vetterli, “The effective rank: A measure of ef- fective dimensionality,” in2007 15th European Signal Processing Conference, 2007, pp. 606–610

    2007

  10. [18]

    Towards automatic assessment of self-supervised speech models using rank,

    Z. Aldeneh, V . Thilak, T. Higuchi, B.-J. Theobald, and T. Likhomanenko, “Towards automatic assessment of self-supervised speech models using rank,”arXiv preprint arXiv:2409.10787, 2024

    2024 arXiv

  11. [19]

    Self- supervised learning with random-projection quantizer for speech recognition,

    C.-C. Chiu, J. Qin, Y . Zhang, J. Yu, and Y . Wu, “Self- supervised learning with random-projection quantizer for speech recognition,” inProceedings of the 39th International Conference on Machine Learning, ser. Proceedings of Machine Learning Research, K. Chaudhuri, S. Jegelk...

    2022

  12. [20]

    Open implementation and study of best-rq for speech processing,

    R. Whetten, T. Parcollet, M. Dinarelli, and Y . Est `eve, “Open implementation and study of best-rq for speech processing,” in Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP) 2024, Workshop on Self-supervision in Audio, Spe...

    2024 arXiv

  13. [21]

    SpeechBrain: A general-purpose speech toolkit,

    M. Ravanelli, T. Parcollet, P. Plantinga, A. Rouhe, S. Cornell, L. Lugosch, C. Subakan, N. Dawalatabad, A. Heba, J. Zhong, J.-C. Chou, S.-L. Yeh, S.-W. Fu, C.-F. Liao, E. Rastorgueva, F. Grondin, W. Aris, H. Na, Y . Gao, R. D. Mori, and Y . Ben- gio, “SpeechBrain: A general-pu...

    2021 arXiv

  14. [22]

    Conformer: Convolution-augmented transformer for speech recognition,

    A. Gulati, J. Qin, C.-C. Chiu, N. Parmar, Y . Zhang, J. Yu, W. Han, S. Wang, Z. Zhang, Y . Wu, and R. Pang, “Conformer: Convolution-augmented transformer for speech recognition,” in Interspeech 2020, 2020, pp. 5036–5040

    2020

  15. [23]

    Some methods for classification and analysis of multivariate observations,

    J. MacQueen, “Some methods for classification and analysis of multivariate observations,” Proc. 5th Berkeley Symp. Math. Stat. Probab., Univ. Calif. 1965/66, 1, 281-297 (1967)., 1967

    1967

  16. [24]

    Least squares quantization in pcm,

    S. Lloyd, “Least squares quantization in pcm,”IEEE Transactions on Information Theory, vol. 28, no. 2, pp. 129–137, 1982

    1982

  17. [25]

    Web-scale k-means clustering,

    D. Sculley, “Web-scale k-means clustering,” inProceedings of the 19th International Conference on World Wide Web, ser. WWW ’10. New York, NY , USA: Association for Computing Machinery, 2010, p. 1177–1178. [Online]. Available: https://doi.org/10.1145/1772690.1772862

    2010

  18. [26]

    Scikit-learn: Machine learning in Python,

    F. Pedregosa, G. Varoquaux, A. Gramfort, V . Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V . Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duchesnay, “Scikit-learn: Machine learning in Python,” Journal of Machine Lear...

    2011

  19. [27]

    k-means++: the advantages of careful seeding,

    D. Arthur and S. Vassilvitskii, “k-means++: the advantages of careful seeding,” inProceedings of the Eighteenth Annual ACM-SIAM Symposium on Discrete Algorithms, ser. SODA ’07. USA: Society for Industrial and Applied Mathematics, 2007, p. 1027–1035

    2007

  20. [28]

    A cluster separation measure,

    D. L. Davies and D. W. Bouldin, “A cluster separation measure,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. PAMI-1, no. 2, pp. 224–227, 1979

    1979

  21. [29]

    Lib- rispeech: an asr corpus based on public domain audio books,

    V . Panayotov, G. Chen, D. Povey, and S. Khudanpur, “Lib- rispeech: an asr corpus based on public domain audio books,” in2015 IEEE international conference on acoustics, speech and signal processing (ICASSP). IEEE, 2015, pp. 5206–5210

    2015

  22. [30]

    wav2vec 2.0: A framework for self-supervised learning of speech repre- sentations,

    A. Baevski, Y . Zhou, A. Mohamed, and M. Auli, “wav2vec 2.0: A framework for self-supervised learning of speech repre- sentations,”Advances in neural information processing systems, vol. 33, pp. 12 449–12 460, 2020

    2020

  23. [31]

    Hubert: Self-supervised speech represen- tation learning by masked prediction of hidden units,

    W.-N. Hsu, B. Bolte, Y .-H. H. Tsai, K. Lakhotia, R. Salakhutdi- nov, and A. Mohamed, “Hubert: Self-supervised speech represen- tation learning by masked prediction of hidden units,”IEEE/ACM transactions on audio, speech, and language processing, vol. 29, pp. 3451–3460, 2021

    2021

  24. [32]

    ContextNet: Improving Convolutional Neural Networks for Automatic Speech Recognition with Global Context,

    W. Han, Z. Zhang, Y . Zhang, J. Yu, C.-C. Chiu, J. Qin, A. Gu- lati, R. Pang, and Y . Wu, “ContextNet: Improving Convolutional Neural Networks for Automatic Speech Recognition with Global Context,” inProc. Interspeech 2020, 2020, pp. 3610–3614

    2020

  25. [33]

    V oxceleb: A large- scale speaker identification dataset,

    A. Nagrani, J. S. Chung, and A. Zisserman, “V oxceleb: A large- scale speaker identification dataset,” inInterspeech 2017, 2017, pp. 2616–2620

    2017

  26. [34]

    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,” inInterspeech 2020, 2020, pp. 3830–3834

    2020

  27. [35]

    Layer-wise analysis of a self-supervised speech representation model,

    A. Pasad, J.-C. Chou, and K. Livescu, “Layer-wise analysis of a self-supervised speech representation model,” in2021 IEEE Auto- matic Speech Recognition and Understanding Workshop (ASRU), 2021, pp. 914–921

    2021

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.