Pith. sign in

REVIEW 3 major objections 5 minor 32 references

Identifying Speaker Information in Feed-Forward Layers of Self-Supervised Speech Transformers

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper claims that speaker information in self-supervised speech Transformers is concentrated in specific feed-forward neurons, which can be found without labels by correlating activations with k-means clusters of speech features and…

desk verdict A clean, label-free extension of the authors' property-neuron work, but the pruning evidence does not yet isolate whether the identified neurons specifically encode speaker information. read the letter →

arxiv 2506.21712 v1 pith:BCIZB5QE submitted 2025-06-26 cs.CL cs.SDeess.AS

Tzu-Quan Lin , Hsi-Chun Cheng , Hung-yi Lee , Hao Tang This is my paper
classification cs.CLcs.SDeess.AS
keywords self-supervisedspeechspeakeridentificationfeed-forwardlayersneuronpruningi-vectorsk-meansclusteringpropertyneuronsone-shot
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

This paper claims that speaker information in self-supervised speech Transformers is concentrated in a small set of feed-forward neurons, and that those neurons can be found without any labels. The method clusters self-supervised features and i-vectors, treats the clusters as coarse phonetic and gender classes, and marks the neurons that co-activate with a cluster as speaker-relevant. Protecting those neurons during one-shot pruning keeps speaker-identification accuracy close to the unpruned model after roughly 70% of the feed-forward parameters are removed, and matches or beats a supervised method that uses phone and gender labels. The result matters because it gives a label-free way to locate where speaker identity lives in these models and shows that pruning can serve as an analysis tool rather than only a compression trick.

What carries the argument

The load-bearing object is the activation-pattern correlation test: each frame is binarized into neurons active in the top 1% (Eq. 1), and a neuron belongs to a cluster's set if it is active in more than 1% of that cluster's frames (Eqs. 2 and 4). Overlapping neurons across clusters are removed (Eqs. 3 and 9), giving the protected sets $P_{ssl}$ and $P_{ive}$. The k-means clusters themselves are the label-free proxy: three SSL clusters approximate vowel/consonant classes, and two i-vector clusters approximate gender, so the identified neurons capture both phonetic and speaker-level structure. During pruning, these sets are exempted from $\ell^1$-norm-based removal of the 3072 hidden dimensions, which is what lets the paper attribute the preserved speaker-identification accuracy to the identified neurons.

What would settle it

Run one-shot pruning on the same models while protecting an equal number of randomly chosen neurons in the same layers, matched for layer and set size; if random protection recovers the same speaker-identification accuracy as the discovered cluster neurons, the cluster-conditioned selection is not what preserves the information. A complementary check is to prune away the discovered neurons and keep everything else: the central claim predicts a sharp speaker-ID collapse.

Watch

Extended reading notes

Core claim

The central claim is that the sets the paper calls SSL cluster neurons and i-vector cluster neurons — dimensions of the first feed-forward layer whose top-1% activations co-occur with a k-means cluster of self-supervised features or of i-vectors — carry a substantial share of the speaker information. With three self-supervised clusters and two i-vector clusters, the clusters line up with broad phonetic classes and with gender (one i-vector cluster is 97.23% male, the other 89.82% female), so the identified neurons act as speaker-relevant units discovered without labels. On SUPERB speaker identification, pruning the feed-forward layers while protecting these neurons recovers most of the loss that regular pruning causes, and in the one-shot setting the protection beats the supervised property-neuron baseline on both MelHuBERT and wav2vec 2.0. The paper further reports that removing either cluster-neuron type from the protected set degrades performance, with the SSL cluster neurons more critical, and that too many clusters leave too few neurons to protect.

Load-bearing premise

The whole identification rests on assuming that a neuron whose top-1% activations co-occur with a k-means cluster more than 1% of the time is one of a few neurons that carry speaker information; if the information is spread thinly across many neurons or lives in activation magnitudes, the protected sets could be arbitrary.

Editorial extensions

If this is right

  • Pruning can remove about 70% of the feed-forward parameters in a single step without a large speaker-identification drop, provided the cluster neurons are spared.
  • The unsupervised identification rivals the supervised property-neuron approach, and in one-shot pruning it beats it on speaker identification for both MelHuBERT (62.96 vs. 58.27 accuracy) and wav2vec 2.0 (70.67 vs. 66.05).
  • Both the SSL cluster neurons and the i-vector cluster neurons are necessary; dropping either from the protected set lowers speaker-identification accuracy, and the SSL cluster neurons are the more critical of the two.
  • The benefit is not limited to speaker tasks: protecting cluster neurons also helps emotion recognition and intent classification under one-shot pruning, sometimes surpassing the unpruned model.
  • Using too many clusters backfires because each additional cluster shrinks the number of identified neurons in middle layers, leaving too few units to protect.

Reading between the lines

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

  • The same unsupervised recipe could be pointed at other properties — emotion, language, noise condition — by clustering embeddings suited to those properties, giving a label-free route to locate task-relevant neurons in any self-supervised speech model.
  • Because SSL cluster neurons matter more than i-vector cluster neurons for speaker identification, the paper's picture suggests speaker identity is partly carried by phonetic structure; a testable prediction is that the protected set should also transfer across languages or recording conditions if it captures speaker rather than corpus-specific units.
  • A matched-control test the paper does not run: protecting an equal number of randomly chosen neurons in the same layers. If random sets preserve speaker identification as well, the cluster-conditioned selection would not be doing the work.
  • The one-shot results that beat the unpruned model suggest protecting cluster neurons may act as structured regularization; one could test whether the same protection ratio improves generalization on held-out speakers or noisy utterances.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes a label-free method for identifying neurons in the feed-forward layers of self-supervised speech Transformers that are related to speaker information. The method extends the property-neuron framework of Lin et al. [20] by replacing supervised labels (phones, gender) with k-means clusters of SSL features and i-vectors. In their experiments on MelHuBERT and wav2vec 2.0, the authors protect the identified 'SSL cluster neurons' and 'i-vector cluster neurons' during iterative and one-shot pruning, and report that speaker identification (SID) performance is preserved much better than with regular pruning, rivaling the supervised property-neuron approach. They also present ablations studying the number of clusters and alternative definitions of i-vector cluster neurons. The central claim is that these cluster neurons play a significant role in encoding speaker information.

Significance. If the central claim is established, the paper would offer a useful unsupervised tool for interpreting speaker-related representations in SSL speech models, with potential practical benefits for pruning. The experimental design is coherent, and the comparison against regular pruning and the supervised property-neuron baseline is a strength. However, the causal interpretation hinges on the assumption that the identified neurons are special relative to other comparably sized neuron sets; this is currently not tested. The absence of error bars, the post-hoc selection of cluster numbers and layer, and the unvalidated cross-model transfer of cluster assignments all weaken the evidence. These issues are addressable with additional experiments, making the manuscript a candidate for revision rather than outright rejection.

major comments (3)
  1. [IV-B and IV-C (Tables I and II)] The pruning-protection evidence does not currently demonstrate that the identified neurons specifically encode speaker information, because no control matched for neuron count and selection is provided. The authors attribute the advantage of the conditioning-and-intersection definition in Eq. (5) to the fact that it identifies fewer neurons, 'allowing for more efficient pruning that avoids protecting too many neurons' (Sec. IV-C). This makes the reported improvements consistent with a neuron-count or frequently-activated-neuron effect rather than with speaker-specific encoding. I request a control in which an equal number of neurons per layer, matched to the protected sets, is selected either randomly or by permuting cluster labels, and the pruning experiment is repeated with that control. Without such a control, the central causal claim in the abstract and conclusion is underdetermined.
  2. [III and IV-A] Key hyperparameters are selected after seeing the clustering and pruning outcomes, which introduces circularity into the main claim. The paper sets kssl=3, kive=2, and layer 9 intentionally because they produce interpretable clusters (Sec. IV-A) and good pruning results. While the ablation in Table II shows that kssl=39 and kssl=100 degrade performance, it does not establish that the chosen values are not optimized for the reported SID numbers. The authors should either prespecify these choices or provide a robustness study across a range of kssl and kive values that shows the pruning-preservation effect is not confined to a narrowly tuned operating point.
  3. [III and IV-B] The claim that the method identifies speaker information is stronger than is supported by the cluster semantics. The i-vector clusters are shown to correspond to gender classes (97.23% male and 89.82% female purity, Sec. IV-A), so the protected neurons are correlated with a coarse demographic attribute rather than with individual speaker identity. In addition, the same MelHuBERT layer-9 cluster assignments are applied to wav2vec 2.0 features without validating that frame-level cluster alignment transfers across models. The authors should temper the 'speaker information' terminology to 'gender- and phone-correlated information' or provide evidence that the identified neurons capture more than these coarse attributes, e.g., by evaluating speaker verification with multiple speakers per gender.
minor comments (5)
  1. [II and III] There are several typographical errors, including 'thie section', 'self-sueprvised', 'Spefically', 'overlappting', 'paramameters', and 'lowℓ1' (missing space in 'low ℓ1'). These should be corrected.
  2. [II, Eq. (1)] The notation 'v1 > top λ%(v)' is ambiguous; please define precisely whether the threshold is the (100−λ)-th percentile or the value exceeded by exactly λ% of dimensions.
  3. [III] No error bars, confidence intervals, or significance tests are reported for any of the pruning results, despite the fact that several comparisons are close (e.g., MelHuBERT iterative SID: 52.66 vs. 54.10 for property neurons). Reporting multiple seeds or bootstrap intervals would strengthen the conclusions.
  4. [Table I] The one-shot regular-pruning SID result for MelHuBERT (54.08) is higher than the iterative regular-pruning result (51.04), which seems counterintuitive given that iterative pruning normally retains more performance. A brief explanation would help the reader.
  5. [IV-C] The subscript differences in Table II (e.g., '49.19-3.47') are not clearly defined in the text. Please state explicitly that the subscripted number is the difference from the Proposed Method row.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation: pruning-protection evidence is an independent empirical test, not an artifact of definition.

full rationale

The central claim—that neurons correlated with SSL and i-vector clusters encode speaker information—is not equivalent to its inputs by construction. The neuron sets are defined by activation co-occurrence statistics (Eqs. 1–4) with k-means clusters, without using speaker labels or pruning outcomes. The pruning-protection experiment (Table I) is an independent evaluation against the external SUPERB benchmark, compares favorably to regular pruning, and is also compared to the labeled property-neuron baseline [20]; the outcome could have failed, so the result is not forced. The paper inherits the activation-pattern and intersection methodology from the authors' own prior work [20], but that self-citation supplies the analysis tool rather than the conclusion, and the paper directly benchmarks against [20]'s supervised approach. The absence of a same-size random-neuron control and the post-hoc choices of kssl=3, layer 9, and the intersection rule over Alternatives A–C raise a validity confound—the pruning benefit may partly reflect neuron count rather than speaker-specific semantics—but this is an empirical-control concern, not circularity. The identified neuron sets are not defined in terms of the SID outcome, and no equation reduces the reported result to a fitted parameter or to the authors' prior claims.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The method depends on several hand-chosen hyperparameters and inherited assumptions from prior work, but it introduces no new physical or architectural entities. The main parameter risk is the post hoc selection of cluster counts (kssl=3, kive=2) and layer 9, which are chosen partly to make the pruning results work.

free parameters (6)
  • lambda (activation threshold) = 1%
    Hand-chosen threshold for binarizing activations; inherited from prior work [20], not swept or validated.
  • rho (co-occurrence threshold) = 1%
    Hand-chosen threshold for declaring a neuron correlated with a cluster; same value as [20].
  • kssl (number of SSL clusters) = 3
    Chosen to yield broad phonetic clusters and enough identified neurons; the ablation (Table II) shows larger values reduce performance, indicating outcome-driven selection.
  • kive (number of i-vector clusters) = 2
    Chosen so clusters align with gender, which mirrors the gender labels used in [20] and helps the interpretability story.
  • layer index for k-means features = 9 (MelHuBERT)
    Intentionally selected as phonetically prominent; the same frame-wise cluster assignments are reused for wav2vec 2.0, which is an additional assumption.
  • pruning target dimensions = 867 (MelHuBERT), 780 (wav2vec 2.0)
    Set to make one-shot and iterative pruning comparable; directly affects the reported performance numbers.
assumptions (5)
  • domain assumption Feed-forward layers in Transformers act as key-value memories where a small set of neurons can represent a property (Geva et al. 2021; Lin et al. 2024).
    This underpins the whole neuron identification and pruning-protection design; if the key-value memory view is wrong, the method loses its foundation.
  • domain assumption Binarizing a frame's activations by top-lambda and counting co-occurrences with a cluster identifies the neurons that represent that cluster.
    The method of [20] is assumed valid; no independent validation of the binarization/co-occurrence thresholding is given in this paper.
  • domain assumption Pruning a neuron and observing task degradation is a valid probe of whether the neuron encodes information for that task.
    The evaluation treats pruning performance as causal evidence of 'encoding'; this is a functional definition that could be questioned.
  • domain assumption K-means clusters of SSL features correspond to broad phone classes, and i-vector clusters correspond to gender.
    The paper shows empirical correspondence (Figure 2 and cluster purity), but this is an assumed property of the features used for clustering.
  • ad hoc to paper Layer 9 of MelHuBERT is phonetically prominent enough that its cluster assignments transfer to wav2vec 2.0.
    The choice of layer is motivated by interpretability and prior knowledge, not by a formal criterion; no layer sweep is reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Identifying Speaker Information in Feed-Forward Layers of Self-Supervised Speech Transformers." pith.science (2026). https://pith.science/paper/BCIZB5QE

@misc{pith2026250621712,
  author       = {Pith},
  title        = {Pith review of: Identifying Speaker Information in Feed-Forward Layers of Self-Supervised Speech Transformers},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BCIZB5QE}},
  note         = {Machine review of arXiv:2506.21712}
}
read the original abstract

In recent years, the impact of self-supervised speech Transformers has extended to speaker-related applications. However, little research has explored how these models encode speaker information. In this work, we address this gap by identifying neurons in the feed-forward layers that are correlated with speaker information. Specifically, we analyze neurons associated with k-means clusters of self-supervised features and i-vectors. Our analysis reveals that these clusters correspond to broad phonetic and gender classes, making them suitable for identifying neurons that represent speakers. By protecting these neurons during pruning, we can significantly preserve performance on speaker-related task, demonstrating their crucial role in encoding speaker information.

Figures

Figures reproduced from arXiv: 2506.21712 by the authors.

Figure 1
Figure 1. An example showing the set relationships of SSL cluster neurons (left), i-vector cluster neurons (middle), and one of the possible alternatives for [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Amounts of broad phone classes within the clusters produced by [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. The number of i-vector cluster neurons in different layers for different [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

32 extracted references · 29 canonical work pages

  1. [20]

    Property Neurons in Self- Supervised Speech Transformers

    Tzu-Quan Lin et al. “Property Neurons in Self- Supervised Speech Transformers”. In: SLT. 2024, pp. 401–408

    work page 2024

  2. [1]

    Effec- tiveness of Self-Supervised Pre-Training for ASR

    Alexei Baevski and Abdelrahman Mohamed. “Effec- tiveness of Self-Supervised Pre-Training for ASR”. In: ICASSP. IEEE, 2020, pp. 7694–7698

    work page 2020

  3. [2]

    data2vec: A General Framework for Self-supervised Learning in Speech, Vision and Language

    Alexei Baevski et al. “data2vec: A General Framework for Self-supervised Learning in Speech, Vision and Language”. In: ICML. 2022

    work page 2022

  4. [3]

    wav2vec 2.0: A framework for self-supervised learning of speech representations

    Alexei Baevski et al. “wav2vec 2.0: A framework for self-supervised learning of speech representations”. In: Advances in Neural Information Processing Systems . V ol. 33. Curran Associates, Inc., 2020, pp. 12449– 12460

    work page 2020

  5. [4]

    Information-Theoretic Understanding of Population Risk Improvement with Model Com- pression

    Yuheng Bu et al. “Information-Theoretic Understanding of Population Risk Improvement with Model Com- pression”. In: Proceedings of the AAAI Conference on Artificial Intelligence 34.4 (2020), pp. 3300–3307

    work page 2020

  6. [5]

    WavLM: Large-scale self- supervised pre-training for full stack speech process- ing

    Sanyuan Chen et al. “WavLM: Large-scale self- supervised pre-training for full stack speech process- ing”. In: IEEE Journal of Selected Topics in Signal Processing 16.6 (2022), pp. 1505–1518

    work page 2022

  7. [6]

    Large-scale self-supervised speech representation learning for automatic speaker verification

    Zhengyang Chen et al. “Large-scale self-supervised speech representation learning for automatic speaker verification”. In: ICASSP. IEEE, 2022, pp. 6147–6151

    work page 2022

  8. [7]

    Self-Supervised Speech Rep- resentations are More Phonetic than Semantic

    Kwanghee Choi et al. “Self-Supervised Speech Rep- resentations are More Phonetic than Semantic”. In: Interspeech. 2024, pp. 4578–4582

    work page 2024

Show all 32 references
  1. [8]

    What do end-to-end speech models learn about speaker, language and channel information? A layer- wise and neuron-level analysis

    Shammur Absar Chowdhury, Nadir Durrani, and Ahmed Ali. “What do end-to-end speech models learn about speaker, language and channel information? A layer- wise and neuron-level analysis”. In: Computer Speech & Language 83 (2024), p. 101539

    2024

  2. [9]

    An unsupervised autoregressive model for speech representation learning

    Yu-An Chung et al. “An unsupervised autoregressive model for speech representation learning”. In: Inter- speech. 2019, pp. 146–150

    2019

  3. [10]

    Front-End Factor Analysis for Speaker Verification

    Najim Dehak et al. “Front-End Factor Analysis for Speaker Verification”. In: IEEE Transactions on Audio, Speech, and Language Processing 19.4 (2011), pp. 788– 798

    2011

  4. [11]

    Exploring wav2vec 2.0 on speaker verification and language identification

    Zhiyun Fan et al. “Exploring wav2vec 2.0 on speaker verification and language identification”. In: Inter- speech. 2021, pp. 1509–1513

    2021

  5. [12]

    The lottery ticket hypothesis: Finding sparse, trainable neural net- works

    Jonathan Frankle and Michael Carbin. “The lottery ticket hypothesis: Finding sparse, trainable neural net- works”. In: ICLR. 2018

    2018

  6. [13]

    Self-supervised pre-training for attention-based encoder-decoder asr model

    Changfeng Gao et al. “Self-supervised pre-training for attention-based encoder-decoder asr model”. In: TASLP 30 (2022), pp. 1763–1774

    2022

  7. [14]

    Transformer feed-forward layers are key-value memories

    Mor Geva et al. “Transformer feed-forward layers are key-value memories”. In: Proceedings of the 2021 Con- 5 ference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics, 2021, pp. 5484–5495

    2021

  8. [15]

    Learning both Weights and Connec- tions for Efficient Neural Network

    Song Han et al. “Learning both Weights and Connec- tions for Efficient Neural Network”. In: Advances in Neural Information Processing Systems. V ol. 28. Curran Associates, Inc., 2015

    2015

  9. [16]

    HuBERT: Self-supervised speech representation learning by masked prediction of hidden units

    Wei-Ning Hsu et al. “HuBERT: Self-supervised speech representation learning by masked prediction of hidden units”. In: IEEE/ACM transactions on audio, speech, and language processing 29 (2021), pp. 3451–3460

    2021

  10. [17]

    SNIP: Single-shot Network Pruning based on Connection Sensitivity

    Namhoon Lee, Thalaiyasingam Ajanthan, and Philip HS Torr. “SNIP: Single-shot Network Pruning based on Connection Sensitivity”. In: ICLR. 2019

    2019

  11. [18]

    MelHu- BERT: A simplified HuBERT on Mel spectrograms

    Tzu-Quan Lin, Hung-yi Lee, and Hao Tang. “MelHu- BERT: A simplified HuBERT on Mel spectrograms”. In: ASRU. 2023, pp. 1–8

    2023

  12. [19]

    Compressing transformer-based self-supervised models for speech processing

    Tzu-Quan Lin et al. “Compressing transformer-based self-supervised models for speech processing”. In: arXiv preprint arXiv:2211.09949 (2022)

    2022 arXiv

  13. [21]

    DeCoAR 2.0: Deep Contextualized Acoustic Representations with Vector Quantization

    Shaoshi Ling and Yuzong Liu. “DeCoAR 2.0: Deep Contextualized Acoustic Representations with Vector Quantization”. In: arXiv preprint arXiv:2012.06659 (2020)

    2020 arXiv

  14. [22]

    DinoSR: Self-Distillation and Online Clustering for Self-supervised Speech Represen- tation Learning

    Alexander H Liu et al. “DinoSR: Self-Distillation and Online Clustering for Self-supervised Speech Represen- tation Learning”. In: Advances in Neural Information Processing Systems . V ol. 36. Curran Associates, Inc., 2024

    2024

  15. [23]

    TERA: Self-Supervised Learning of Transformer Encoder Rep- resentation for Speech

    Andy T Liu, Shang-Wen Li, and Hung-yi Lee. “TERA: Self-Supervised Learning of Transformer Encoder Rep- resentation for Speech”. In: IEEE/ACM Transactions on Audio, Speech, and Language Processing 29 (2021), pp. 2351–2366

    2021

  16. [24]

    Introducing phonetic information to speaker embedding for speaker verification

    Yi Liu et al. “Introducing phonetic information to speaker embedding for speaker verification”. In: EURASIP Journal on Audio, Speech, and Music Pro- cessing (2019)

    2019

  17. [25]

    Speaker embedding extraction with pho- netic information

    Yi Liu et al. “Speaker embedding extraction with pho- netic information”. In: Interspeech (2018), pp. 2247– 2251

    2018

  18. [26]

    Probing self-supervised speech models for phonetic and phonemic information: a case study in aspiration

    Kinan Martin et al. “Probing self-supervised speech models for phonetic and phonemic information: a case study in aspiration”. In: Interspeech. 2023, pp. 251–255

    2023

  19. [27]

    Towards Supervised Performance on Speaker Verification with Self-Supervised Learning by Leveraging Large-Scale ASR Models

    Victor Miara, Theo Lepage, and Reda Dehak. “Towards Supervised Performance on Speaker Verification with Self-Supervised Learning by Leveraging Large-Scale ASR Models”. In: Interspeech. 2024, pp. 2660–2664

    2024

  20. [28]

    Orthogonality and isotropy of speaker and phonetic information in self-supervised speech representations

    Mukhtar Mohamed et al. “Orthogonality and isotropy of speaker and phonetic information in self-supervised speech representations”. In: Interspeech. 2024, pp. 3625–3629

    2024

  21. [29]

    X-vectors: Robust dnn embeddings for speaker recognition

    David Snyder et al. “X-vectors: Robust dnn embeddings for speaker recognition”. In: 2018 IEEE international conference on acoustics, speech and signal processing (ICASSP). IEEE. 2018, pp. 5329–5333

    2018

  22. [30]

    A fine-tuned wav2vec 2.0/HuBERT benchmark for speech emotion recognition, speaker verification and spoken language understanding

    Yingzhi Wang, Abdelmoumene Boumadane, and Ab- delwahab Heba. “A fine-tuned wav2vec 2.0/HuBERT benchmark for speech emotion recognition, speaker verification and spoken language understanding”. In: arXiv:2111.02735 (2022)

    2022 arXiv

  23. [31]

    Phonetic analysis of self-supervised representations of english speech

    Dan Wells, Hao Tang, and Korin Richmond. “Phonetic analysis of self-supervised representations of english speech”. In: Interspeech. 2022, pp. 3583–3587

    2022

  24. [32]

    SUPERB: Speech processing Universal PERformance Benchmark

    Shu-wen Yang et al. “SUPERB: Speech processing Universal PERformance Benchmark”. In: Interspeech. 2021, pp. 1194–1198. 6

    2021

Pith tools

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