Pith. sign in

REVIEW 3 major objections 5 minor 2 cited by

Comprehensive Layer-wise Analysis of SSL Models for Audio Deepfake Detection

T0 review · 3 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read Audio deepfake detection can drop most of a self-supervised speech model's layers and still match or beat full-model accuracy.

desk verdict A broad, useful ablation of SSL layers for deepfake detection, whose precise layer-count recommendation outruns the reported statistics. read the letter →

arxiv 2502.03559 v2 pith:PGFCI7B5 submitted 2025-02-05 eess.AS cs.SD

Yassine El Kheir , Youness Samih , Suraj Maharjan , Tim Polzehl , Sebastian Möller This is my paper · ORCID
classification eess.AScs.SD
keywords audiodeepfakedetectionself-supervisedlearninglayer-wiseanalysisWav2Vec2HuBERTWavLMmodelcompressionequalerrorrate
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 argues that, across six self-supervised speech models and ten datasets spanning English, Chinese, and Spanish, plus partial, song, and scene fakes, the information that separates real speech from deepfakes is concentrated in the lower transformer layers. It therefore claims that a detector can discard most of a model's upper layers, keeping only the first 4 to 6 layers of small models or 10 to 12 layers of large models, and still match, and often beat, the full model's equal error rate. If right, this makes audio deepfake detection substantially cheaper and faster without sacrificing accuracy, and it reframes what SSL models actually learn for this task: local artifact patterns rather than global utterance abstractions.

What carries the argument

The load-bearing object is the weighted layer-aggregation front end, $h_{\mathrm{final}} = \sum_{l=1}^{L} w_l h_l$, where $h_l$ are the per-layer transformer outputs and the weights $w_l$ are learnable and softmax-normalized. By freezing the SSL front-end and reading the trained weights, the paper obtains an empirical importance score for each layer; by then evaluating truncated models that keep only the first $k$ layers, it converts those scores into a parameter-reduction recipe. The two back-end heads, a two-layer feedforward network with attentive statistical pooling and AASIST, a graph-based audio anti-spoofing classifier, act as controls showing the lower-layer preference is not specific to one classifier.

What would settle it

Train a truncated model, say Wav2Vec2-Small-6, and its full counterpart on one corpus, fix the layer count before looking at test results, and evaluate on a held-out fake-audio generator or language not among the paper's ten datasets; if the full model clearly wins on several such sets, or if the learned layer weights spread toward upper layers, the 'lower layers suffice' claim fails.

Watch

Extended reading notes

Core claim

The central discovery is a layer-wise importance profile for frozen SSL front-ends: learned softmax weights assigned to each transformer layer, averaged over datasets, concentrate in layers 1-6 for small models and 1-12 for large models, with the strongest weights around layers 4-7. When the model is truncated to these lower layers and retrained with either a lightweight feedforward head or the AASIST graph-attention anti-spoofing head, its averaged EER over ten datasets is 20.98% for the best small six-layer configuration versus 21.74% for full small models, and 17.78% for the best large twelve-layer configuration versus 17.97% for full large models. The paper interprets this as evidence that lower layers capture local, sub-band, and short-time artifacts left by speech synthesis and voice conversion, while upper layers carry less task-relevant abstraction.

Load-bearing premise

The recommended layer counts are chosen after seeing results on the same benchmark sets, and the average that drives them is itself pulled hardest by datasets with EERs near 60 percent, so the whole recommendation rests on that post-hoc average being representative of the deepfake world.

Editorial extensions

If this is right

  • Detectors built from the first 4-6 layers of small models or 10-12 layers of large models cut parameter count by at least half while keeping average EER within roughly 0.3-0.7 points of the full model.
  • Lower-layer truncation improves generalization on several benchmarks, for example Wav2Vec2-Small-6 with a feedforward head reaches 0.48% EER on ASVspoof2019 LA and Wav2Vec2-Large-12 with AASIST reaches 0.21%, both beating their full models.
  • The reduction holds consistently across English, Chinese, and Spanish and across full, partial, song, and scene fakes, so it is not an artifact of one language or one attack type.
  • The main exceptions are partial-fake and scene-fake data, where full models still outperform truncated ones on PartialSpoof and SceneFake, though by small margins, indicating that upper layers matter when artifact location or global context is involved.

Reading between the lines

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

  • Editorial extension: the layer-count recommendation is selected on the same ten datasets used for reporting, so a stronger test would fix 4-6 and 10-12 layers beforehand and validate on a new language or a new generator; the paper does not run that held-out protocol.
  • Editorial extension: if the lower-layer preference reflects local synthesis artifacts, then truncated detectors should degrade most on attacks that hide artifacts at longer time scales or in global prosody, and comparing truncation behavior across latest neural codec and audio-language-model generators would test this.
  • Editorial extension: because large models matter most on the hardest recent datasets, the practical recipe may be a small truncated model for low-cost screening plus a large truncated model only when budget allows, rather than one universal layer count.
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

3 major / 5 minor

Summary. The paper presents a layer-wise analysis of six frozen SSL front-ends (Wav2Vec2, Hubert, WavLM in small/large versions) for audio deepfake detection across ten diverse datasets spanning multiple languages and scenarios (full, partial, song, scene). The authors learn softmax-weighted combinations of transformer-layer outputs (Eq. 1-2) with two back-end classifiers (FFN and AASIST), and additionally train truncated models using only the first 2-12 layers. The central claim is that lower layers provide the most discriminative features and that reduced-layer models using 4-6 layers (small) or 10-12 layers (large) achieve comparable or better average EER than the full models. The paper reports mean EER over three seeds for each configuration.

Significance. If the main claim is supported, the paper would have practical value: it quantifies how much of a frozen SSL model can be discarded for deepfake detection, potentially reducing compute and memory substantially. The experimental scope is a strength: six front-ends, two back-ends, and ten datasets spanning languages and deepfake types, with direct truncated-model comparisons rather than only correlational weight analysis. The authors also release code and models. However, the headline quantitative recommendation (4-6 and 10-12 layers) is built on very small average-EER differences computed on the same evaluation sets that were used to choose the ranges, and no variance or significance testing is reported. The qualitative trend that lower layers contain useful discriminative information is plausible and partially confirmed by the truncated models, but the paper's stronger wording ('consistently', 'optimal') is not fully supported by the evidence.

major comments (3)
  1. [§4.2, §5.3, Table 4] The recommended reduced-layer ranges (4-6 for small, 10-12 for large) are selected based on the same evaluation sets used to report final performance, and no error bars, standard deviations, confidence intervals, or significance tests are provided. The reported mean EER over 3 seeds leaves differences such as the 6-layer small A_VGout of 20.98% vs the full model's 21.74%, and the 12-layer large 17.78% vs full 17.97%, within the range of selection noise. In fact, Table 4 shows that for WavLM-Large the full model is 0.09 points better than the 12-layer version, and for several model/back-end pairs (e.g., Wav2Vec-Small-AASIST, WavLM-Large-AASIST) the full model is the best configuration overall. The average EER is dominated by the hardest datasets (ADD23, with EERs around 45-75%), while easier datasets like LA19 have EERs near 0.5%, so the pooled average may not reflect per-dataset behavior. The qualitative conclusion that truncated models are broadly comparable may survive, but the precise optimal-layer recommendation and the word 'consistently' require a hold-out evaluation or at least bootstrap confidence intervals over datasets and seeds.
  2. [Abstract, §5.2] The abstract states that 'lower layers consistently provide the most discriminative features,' but the paper's own results contain notable exceptions. For SceneFake, the full models for both Wav2Vec-Small and WavLM-Small consistently outperform their reduced-layer versions (by about 0.5-1.5% EER), and for PartialSpoof the full models consistently outperform reduced-layer models, as acknowledged in §5.2. In Table 4, Wav2Vec2-Large-FFN and WavLM-Large-AASIST full models achieve the best A_VGin. The qualitative finding that several reduced-layer configurations match or beat the full model on multiple datasets is valuable, but the claim of universal consistency should be softened to 'often' or 'in most datasets', and the exceptions should be explicitly factored into the conclusion.
  3. [§3, §5.1, Figures 2-3] The layer-importance analysis uses the learned softmax weights w_l from Eq. (2) to conclude that lower layers are 'more important.' However, these weights are not identifiable as causal contributions: the outputs of different transformer layers are highly correlated, and the weights are jointly optimized with a classifier, so any linear recombination of correlated features can produce many equivalent weight distributions. The truncated-model experiments in §5.2 provide a more direct measure of the value of lower layers, and they are the strongest evidence in the paper. The weight-based analysis should be presented as a heuristic indicator rather than a direct measurement of layer importance, or supplemented with an ablation that removes individual layers (e.g., leave-one-out) to confirm the interpretation.
minor comments (5)
  1. [Figure 2, Figure 3] The caption text 'A VERAGE' contains an unintended space; it should read 'AVERAGE'.
  2. [Tables 1 and 3 and text] There are repeated typos in model names and back-end names: 'Wav2Vev2' instead of 'Wav2Vec2', 'ASSIST' instead of 'AASIST', and 'Wav2Vec-Small' vs 'Wav2Vec2-Small' are used inconsistently.
  3. [§5.3, Table 4] The notation A_VGin and A_VGout is confusing: A_VGin is the average over datasets within a model, and A_VGout is the average over models, but the subscripts 'in' and 'out' are not intuitive. Please define these more explicitly in the text and table caption.
  4. [§6] The sentence 'Tables 1 and 3 show a10.95% difference' is missing a space after 'a'; rephrase to avoid ambiguity (e.g., 'a difference of 10.95%').
  5. [§4.2] The paper states 'Each experiment is repeated 3 times with different seeds for reliable results and the average results are reported,' but no standard deviations or variance measures are given anywhere. Reporting at least the standard deviation of EER over the three seeds would make the reliability of the small average differences assessable.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the layer-importance and reduced-layer claims are empirical results measured on external benchmarks, not consequences of the model definitions.

full rationale

The paper's central claims are supported by two independent sets of measurements. First, the layer-wise importance weights in Equation 1 are learned parameters (softmax-normalized, initialized to ones) trained to minimize cross-entropy on each dataset; the resulting heatmaps in Figures 2 and 3 are observed outcomes, not identities. The conclusion that lower layers receive higher weights is an empirical finding, and the possibility that correlated layer representations affect weight interpretation is a methodological limitation, not a circular reduction. Second, the reduced-layer experiments in Section 5.2 and Tables 1, 3, and 4 physically truncate the frozen SSL front-end to the first X layers, retrain the back-end classifier, and evaluate on external benchmark datasets (LA19, LA21, DF21, ADD23, HABLA, PartialSpoof, HAD, CtrSVDD, SceneFake). These are measured EER values, not quantities forced by construction. The recommended ranges of 4-6 layers for small models and 10-12 layers for large models are selected after inspecting these same results, which is a post-hoc model-selection concern that could inflate perceived performance or overstate consistency; however, that is a correctness/statistical risk rather than a definitional circularity. The paper does not define layer importance in terms of the final EER, does not fit a parameter and then rename it as a prediction, and does not rely on any load-bearing self-citation. The GitHub release of trained models and code further supports reproducibility. Overall, the derivation chain is self-contained against external benchmarks, and no specific reduction of a claimed result to its own inputs can be exhibited.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The paper's central claim is empirical. The main fitted quantities are the layer weights and the post-hoc chosen layer counts. No new physical entities are introduced. The key axioms are the validity of the weighted-sum feature aggregation and the EER averaging across heterogeneous datasets.

free parameters (2)
  • Layer importance weights w_l = not reported per layer in text
    Learned softmax weights in Equation 2, fitted on each dataset, used to infer which layers matter. They are the object of analysis, not a theoretical constant.
  • Reduced layer counts (4-6 small, 10-12 large) = 4, 6 (small); 10, 12 (large)
    Selected post-hoc as 'optimal' based on the same evaluation datasets; a model-selection choice that affects the headline claim.
assumptions (3)
  • ad hoc to paper A weighted sum of layer outputs (Equation 1) with softmax weights is a sufficient aggregation for deepfake classification.
    The entire layer-contribution analysis assumes these weights faithfully represent each layer's contribution; no justification beyond it being a learnable combination.
  • domain assumption Frozen SSL features from each transformer layer are meaningful inputs for the downstream classifier.
    The method freezes the SSL models and trains only the back-end, assuming the pre-trained representations are informative for deepfake detection.
  • ad hoc to paper Mean EER across datasets is a valid aggregate for selecting model configurations.
    Table 4 averages EER over datasets with very different scales (0.5% to 60%), equal-weighting each dataset; this aggregation drives the layer-count recommendation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Comprehensive Layer-wise Analysis of SSL Models for Audio Deepfake Detection." pith.science (2026). https://pith.science/paper/PGFCI7B5

@misc{pith2026250203559,
  author       = {Pith},
  title        = {Pith review of: Comprehensive Layer-wise Analysis of SSL Models for Audio Deepfake Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PGFCI7B5}},
  note         = {Machine review of arXiv:2502.03559}
}
read the original abstract

This paper conducts a comprehensive layer-wise analysis of self-supervised learning (SSL) models for audio deepfake detection across diverse contexts, including multilingual datasets (English, Chinese, Spanish), partial, song, and scene-based deepfake scenarios. By systematically evaluating the contributions of different transformer layers, we uncover critical insights into model behavior and performance. Our findings reveal that lower layers consistently provide the most discriminative features, while higher layers capture less relevant information. Notably, all models achieve competitive equal error rate (EER) scores even when employing a reduced number of layers. This indicates that we can reduce computational costs and increase the inference speed of detecting deepfakes by utilizing only a few lower layers. This work enhances our understanding of SSL models in deepfake detection, offering valuable insights applicable across varied linguistic and contextual settings. Our trained models and code are publicly available: https://github.com/Yaselley/SSL_Layerwise_Deepfake.

Figures

Figures reproduced from arXiv: 2502.03559 by the authors.

Figure 1
Figure 1. Layer-wise Contribution Framework. The framework consists of SSL models as front-ends to extract features and a back-end classifier. The front-end SSL models remain frozen during the experiments to evaluate the layer-wise feature contribution. et al., 2024), our study offers a broader perspective. Our methodology uniquely extends beyond these limitations by comprehensively assessing deepfake detection tasks across v… view at source ↗
Figure 2
Figure 2. Heatmap of Normalized Layer-wise weights Across Various Datasets using Small SSL Models. AVERAGE [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Heatmap of Normalized Layer-wise weights Across Various Datasets using Large SSL Models. AVERAGE [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Multi-level SSL Feature Gating for Audio Deepfake Detection

    cs.SD 2025-09 conditional novelty 5.0 of 10

    An XLS-R based audio deepfake detector combining gated multi-kernel convolutions with a CKA dissimilarity loss reports top EERs on 19LA, 21DF, and In-The-Wild benchmarks.

  2. Two Views, One Truth: Spectral and Self-Supervised Features Fusion for Robust Speech Deepfake Detection

    cs.SD 2025-07 conditional novelty 4.0 of 10

    Fusing CQCC spectral features with Wav2Vec2.0 embeddings via cross-attention lowers average equal error rate from 10.87% to 6.80% across four speech deepfake benchmarks.

Reference graph

Works this paper leans on

59 extracted references · 30 canonical work pages · cited by 2 Pith papers

  1. [1]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block STRING...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Federico Alegre, Asmaa Amehraye, and Nicholas Evans. 2013. A one-class classification approach to generalised speaker verification spoofing countermeasures using local binary patterns. In 2013 IEEE Sixth International Conference on Biometrics: Theory, Applications and Systems (BTAS), pages 1--8. IEEE

    work page 2013

  4. [4]

    Z Khamar Anjum and R Kumara Swamy. 2017. Spoofing and countermeasures for speaker verification: A review. In 2017 International Conference on Wireless Communications, Signal Processing and Networking (WiSPNET), pages 467--471. IEEE

    work page 2017

  5. [5]

    Arun Babu, Changhan Wang, Andros Tjandra, Kushal Lakhotia, Qiantong Xu, Naman Goyal, Kritika Singh, Patrick Von Platen, Yatharth Saraf, Juan Pino, et al. 2021. Xls-r: Self-supervised cross-lingual speech representation learning at scale. arXiv preprint arXiv:2111.09296

    arXiv 2021

  6. [6]

    Alexei Baevski, Yuhao Zhou, Abdelrahman Mohamed, and Michael Auli. 2020. wav2vec 2.0: A framework for self-supervised learning of speech representations. Advances in neural information processing systems, 33:12449--12460

    2020

  7. [7]

    Emily R Bartusiak and Edward J Delp. 2022. Transformer-based speech synthesizer attribution in an open set scenario. In 2022 21st IEEE International Conference on Machine Learning and Applications (ICMLA), pages 329--336. IEEE

    work page 2022

  8. [8]

    Lasse Borgholt, Jakob Drachmann Havtorn, Joakim Edin, Lars Maal e, and Christian Igel. 2022. A brief overview of unsupervised neural speech representation learning. arXiv preprint arXiv:2203.01829

    arXiv 2022

Show all 59 references
  1. [9]

    Sandipan Chakroborty, Anindya Roy, and Goutam Saha. 2008. Improved closed set text-independent speaker identification by combining mfcc with evidence from flipped filter banks. International Journal of Electronics and Communication Engineering, 2(11):2554--2561

    2008

  2. [10]

    Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, et al. 2022. Wavlm: Large-scale self-supervised pre-training for full stack speech processing. IEEE Journal of Selected Topics in Signal Processing, ...

    2022

  3. [11]

    Jesse Damiani. 2019. A voice deepfake was used to scam a ceo out of \ 243,000. Forbes Magazine

    2019

  4. [12]

    Habla: A dataset of latin american spanish accents for voice anti-spoofing

    Pablo Andr \'e s Tamayo Fl \'o rez, Rub \'e n Manrique, and Bernardo Pereira Nunes. Habla: A dataset of latin american spanish accents for voice anti-spoofing

  5. [13]

    Yinlin Guo, Haofan Huang, Xi Chen, He Zhao, and Yuehai Wang. 2024. Audio deepfake detection with self-supervised wavlm and multi-fusion attentive classifier. In ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 12702--12706. IEEE

    2024

  6. [14]

    Anmol Guragain, Tianchi Liu, Zihan Pan, Hardik B Sailor, and Qiongqiong Wang. 2024. Speech foundation model ensembles for the controlled singing voice deepfake detection (ctrsvdd) challenge 2024. arXiv preprint arXiv:2409.02302

    2024 arXiv

  7. [15]

    Wei-Ning Hsu, Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal Lakhotia, Ruslan Salakhutdinov, and Abdelrahman Mohamed. 2021. Hubert: Self-supervised speech representation learning by masked prediction of hidden units. IEEE/ACM transactions on audio, speech, and language processin...

    2021

  8. [16]

    Wen-Chin Huang, Lester Phillip Violeta, Songxiang Liu, Jiatong Shi, and Tomoki Toda. 2023. The singing voice conversion challenge 2023. In 2023 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU), pages 1--8. IEEE

    2023

  9. [17]

    Jee-weon Jung, Hee-Soo Heo, Hemlata Tak, Hye-jin Shim, Joon Son Chung, Bong-Jin Lee, Ha-Jin Yu, and Nicholas Evans. 2022. Aasist: Audio anti-spoofing using integrated spectro-temporal graph attention networks. In ICASSP 2022-2022 IEEE international conference on acoustics, spe...

    2022

  10. [18]

    Awais Khan, Khalid Mahmood Malik, and Shah Nawaz. 2024. Frame-to-utterance convergence: A spectra-temporal approach for unified spoofing detection. In ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 10761--10765. IEEE

    2024

  11. [19]

    Awais Khan, Khalid Mahmood Malik, James Ryan, and Mikul Saravanan. 2022. Voice spoofing countermeasures: Taxonomy, state-of-the-art, experimental analysis of generalizability, open challenges, and the way forward. arXiv preprint arXiv:2210.00417

    2022 arXiv

  12. [20]

    Yogesh Kumar, Apeksha Koul, and Chamkaur Singh. 2023. A deep learning approaches in text-to-speech system: a systematic review and recent research perspective. Multimedia Tools and Applications, 82(10):15171--15197

    2023

  13. [21]

    Cheng-I Lai, Nanxin Chen, Jes \'u s Villalba, and Najim Dehak. 2019. Assert: Anti-spoofing with squeeze-excitation and residual networks. arXiv preprint arXiv:1904.01120

    2019 arXiv

  14. [22]

    Galina Lavrentyeva, Sergey Novoselov, Egor Malykh, Alexander Kozlov, Oleg Kudashev, and Vadim Shchemelinin. 2017. Audio replay attack detection with deep learning frameworks. In Interspeech, pages 82--86

    2017

  15. [23]

    Jin Woo Lee, Eungbeom Kim, Junghyun Koo, and Kyogu Lee. 2022. Representation selective self-distillation and wav2vec 2.0 feature exploration for spoof-aware speaker verification. arXiv preprint arXiv:2204.02639

    2022 arXiv

  16. [24]

    Zhenchun Lei, Yingen Yang, Changhong Liu, and Jihua Ye. 2020. Siamese convolutional neural network using gaussian probability feature for spoofing speech detection. In Interspeech, pages 1116--1120

    2020

  17. [25]

    Menglu Li, Yasaman Ahmadiadli, and Xiao-Ping Zhang. 2024. Audio anti-spoofing detection: A survey. arXiv preprint arXiv:2404.13914

    2024 arXiv

  18. [26]

    Wei Li, Jichen Yang, and Pei Lin. 2023. Investigation of the influence of blocks on the linear spectrum for synthetic speech detection. Electronics Letters, 59(9):e12797

    2023

  19. [27]

    Tianchi Liu, Lin Zhang, Rohan Kumar Das, Yi Ma, Ruijie Tao, and Haizhou Li. 2024. How do neural spoofing countermeasures detect partially spoofed audio? arXiv preprint arXiv:2406.02483

    2024 arXiv

  20. [28]

    Juan M Mart \' n-Do \ n as and Aitor \'A lvarez. 2022. The vicomtech audio deepfake detection system based on wav2vec2 for the 2022 add challenge. In ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 9241--9245. IEEE

    2022

  21. [29]

    Abdelrahman Mohamed, Hung-yi Lee, Lasse Borgholt, Jakob D Havtorn, Joakim Edin, Christian Igel, Katrin Kirchhoff, Shang-Wen Li, Karen Livescu, Lars Maal e, et al. 2022. Self-supervised speech representation learning: A review. IEEE Journal of Selected Topics in Signal Processi...

    2022

  22. [30]

    Andreas Nautsch, Xin Wang, Nicholas Evans, Tomi H Kinnunen, Ville Vestman, Massimiliano Todisco, H \'e ctor Delgado, Md Sahidullah, Junichi Yamagishi, and Kong Aik Lee. 2021. Asvspoof 2019: spoofing countermeasures for the detection of synthesized, converted and replayed speec...

    2021

  23. [31]

    Koji Okabe, Takafumi Koshinaka, and Koichi Shinoda. 2018. Attentive statistics pooling for deep speaker embedding. arXiv preprint arXiv:1803.10963

    2018 arXiv

  24. [32]

    Zihan Pan, Tianchi Liu, Hardik B Sailor, and Qiongqiong Wang. 2024. Attentive merging of hidden embeddings from pre-trained speech model for anti-spoofing detection. arXiv preprint arXiv:2406.10283

    2024 arXiv

  25. [33]

    Mirco Ravanelli and Yoshua Bengio. 2018. Speaker recognition from raw waveform with sincnet. In 2018 IEEE spoken language technology workshop (SLT), pages 1021--1028. IEEE

    2018

  26. [34]

    Sahidullah, Tomi Kinnunen, and Cemal Hanilçi

    Md. Sahidullah, Tomi Kinnunen, and Cemal Hanilçi. 2015. https://doi.org/10.21437/Interspeech.2015-472 A comparison of features for synthetic speech detection . In Proc. Interspeech 2015, pages 2087--2091

    2015 doi

  27. [35]

    Yao Shi, Hui Bu, Xin Xu, Shaoji Zhang, and Ming Li. 2020. Aishell-3: A multi-speaker mandarin tts corpus and the baselines. arXiv preprint arXiv:2010.11567

    2020 arXiv

  28. [36]

    Yi-Jen Shih, Zoi Gkalitsiou, Alexandros G Dimakis, and David Harwath. 2024. Self-supervised speech models for word-level stuttered speech detection. arXiv preprint arXiv:2409.10704

    2024 arXiv

  29. [37]

    Hye-jin Shim, Jungwoo Heo, Jae-Han Park, Ga-Hui Lee, and Ha-Jin Yu. 2022. Graph attentive feature aggregation for text-independent speaker verification. In ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 7972--7976. IEEE

    2022

  30. [38]

    Suwon Shon, Ankita Pasad, Felix Wu, Pablo Brusco, Yoav Artzi, Karen Livescu, and Kyu J Han. 2022. Slue: New benchmark tasks for spoken language understanding evaluation on natural speech. In ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Process...

    2022

  31. [39]

    Kaavya Sriskandaraja, Vidhyasaharan Sethu, Phu Ngoc Le, and Eliathamby Ambikairajah. 2016. Investigation of sub-band discriminative information between spoofed and genuine speech. In Interspeech, pages 1710--1714

    2016

  32. [40]

    Hemlata Tak, Jose Patino, Andreas Nautsch, Nicholas Evans, and Massimiliano Todisco. 2020. An explainability study of the constant q cepstral coefficient spoofing countermeasure for automatic speaker verification. arXiv preprint arXiv:2004.06422

    2020 arXiv

  33. [41]

    Hemlata Tak, Jose Patino, Massimiliano Todisco, Andreas Nautsch, Nicholas Evans, and Anthony Larcher. 2021. End-to-end anti-spoofing with rawnet2. In ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 6369--6373. IEEE

    2021

  34. [42]

    Hemlata Tak, Massimiliano Todisco, Xin Wang, Jee-weon Jung, Junichi Yamagishi, and Nicholas Evans. 2022. Automatic speaker verification spoofing and deepfake detection using wav2vec 2.0 and data augmentation. arXiv preprint arXiv:2202.12233

    2022 arXiv

  35. [43]

    Massimiliano Todisco, H \'e ctor Delgado, and Nicholas Evans. 2017. Constant q cepstral coefficients: A spoofing countermeasure for automatic speaker verification. Computer Speech & Language, 45:516--535

    2017

  36. [44]

    Anton Tomilov, Aleksei Svishchev, Marina Volkova, Artem Chirkovskiy, Alexander Kondratev, and Galina Lavrentyeva. 2021. Stc antispoofing systems for the asvspoof2021 challenge. In Proc. ASVspoof 2021 Workshop, pages 61--67

    2021

  37. [45]

    Hsiang-Sheng Tsai, Heng-Jui Chang, Wen-Chin Huang, Zili Huang, Kushal Lakhotia, Shu-wen Yang, Shuyan Dong, Andy T Liu, Cheng-I Jeff Lai, Jiatong Shi, et al. 2022. Superb-sg: Enhanced speech processing universal performance benchmark for semantic and generative capabilities. ar...

    2022 arXiv

  38. [46]

    Xin Wang and Junichi Yamagishi. 2021. Investigating self-supervised front ends for speech spoofing countermeasures. arXiv preprint arXiv:2111.07725

    2021 arXiv

  39. [47]

    Zhiyong Wang, Ruibo Fu, Zhengqi Wen, Jianhua Tao, Xiaopeng Wang, Yuankun Xie, Xin Qi, Shuchen Shi, Yi Lu, Yukun Liu, et al. 2024. Mixture of experts fusion for fake audio detection using frozen wav2vec 2.0. arXiv preprint arXiv:2409.11909

    2024 arXiv

  40. [48]

    Yang Xie, Zhenchuan Zhang, and Yingchun Yang. 2021. Siamese network with wav2vec feature for spoofing speech detection. In Interspeech, pages 4269--4273

    2021

  41. [49]

    Yuankun Xie, Yi Lu, Ruibo Fu, Zhengqi Wen, Zhiyong Wang, Jianhua Tao, Xin Qi, Xiaopeng Wang, Yukun Liu, Haonan Cheng, Long Ye, and Yi Sun. 2024. https://arxiv.org/abs/2405.04880 The codecfake dataset and countermeasures for the universally detection of deepfake audio . Preprin...

    2024 arXiv

  42. [50]

    Junichi Yamagishi, Xin Wang, Massimiliano Todisco, Md Sahidullah, Jose Patino, Andreas Nautsch, Xuechen Liu, Kong Aik Lee, Tomi Kinnunen, Nicholas Evans, et al. 2021. Asvspoof 2021: accelerating progress in spoofed and deepfake speech detection. In ASVspoof 2021 Workshop-Autom...

    2021

  43. [51]

    Shu-wen Yang, Po-Han Chi, Yung-Sung Chuang, Cheng-I Jeff Lai, Kushal Lakhotia, Yist Y Lin, Andy T Liu, Jiatong Shi, Xuankai Chang, Guan-Ting Lin, et al. 2021. Superb: Speech processing universal performance benchmark. arXiv preprint arXiv:2105.01051

    2021 arXiv

  44. [52]

    Jiangyan Yi, Ye Bai, Jianhua Tao, Haoxin Ma, Zhengkun Tian, Chenglong Wang, Tao Wang, and Ruibo Fu. 2021. Half-truth: A partially fake audio detection dataset. arXiv preprint arXiv:2104.03617

    2021 arXiv

  45. [53]

    Jiangyan Yi, Ruibo Fu, Jianhua Tao, Shuai Nie, Haoxin Ma, Chenglong Wang, Tao Wang, Zhengkun Tian, Ye Bai, Cunhang Fan, et al. 2022. Add 2022: the first audio deep synthesis detection challenge. In ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal ...

    2022

  46. [54]

    Jiangyan Yi, Jianhua Tao, Ruibo Fu, Xinrui Yan, Chenglong Wang, Tao Wang, Chu Yuan Zhang, Xiaohui Zhang, Yan Zhao, Yong Ren, et al. 2023 a . Add 2023: the second audio deepfake detection challenge. arXiv preprint arXiv:2305.13774

    2023 arXiv

  47. [55]

    Jiangyan Yi, Chenglong Wang, Jianhua Tao, Chu Yuan Zhang, Cunhang Fan, Zhengkun Tian, Haoxin Ma, and Ruibo Fu. 2024. Scenefake: An initial dataset and benchmarks for scene fake audio detection. Pattern Recognition, 152:110468

    2024

  48. [56]

    Jiangyan Yi, Chenglong Wang, Jianhua Tao, Xiaohui Zhang, Chu Yuan Zhang, and Yan Zhao. 2023 b . Audio deepfake detection: A survey. arXiv preprint arXiv:2308.14970

    2023 arXiv

  49. [57]

    Lin Zhang, Xin Wang, Erica Cooper, Junichi Yamagishi, Jose Patino, and Nicholas Evans. 2021. An initial investigation for detecting partially spoofed audio. arXiv preprint arXiv:2104.02518

    2021 arXiv

  50. [58]

    You Zhang, Yongyi Zang, Jiatong Shi, Ryuichi Yamamoto, Jionghao Han, Yuxun Tang, Tomoki Toda, and Zhiyao Duan. 2024. Svdd challenge 2024: A singing voice deepfake detection challenge evaluation plan. arXiv preprint arXiv:2405.05244

    2024 arXiv

  51. [59]

    Yuxiang Zhang12, Wenchao Wang12, and Pengyuan Zhang12. 2021. The effect of silence and dual-band fusion in anti-spoofing system. In Proc. Interspeech

    2021

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.