Pith. sign in

REVIEW 5 major objections 6 minor 33 references

Phoneme-Level Feature Discrepancies: A Key to Detecting Sophisticated Speech Deepfakes

T0 review · 5 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Deepfake speech carries phoneme-level inconsistencies that a pooling-plus-graph detector can catch, achieving 7.12% EER on ASVspoof2021 DF.

desk verdict Strong cross-dataset deepfake detector, but the phoneme-discrepancy mechanism is unproven while the recognizer's 0.4 PER confound is untested. read the letter →

arxiv 2412.12619 v1 pith:CTY6PAVJ submitted 2024-12-17 cs.SD cs.AIeess.AS

Kuiyuan Zhang , Zhongyun Hua , Rushi Lan , Yushu Zhang , Yifang Guo This is my paper · ORCID
classification cs.SDcs.AIeess.AS
keywords speechdeepfakedetectionphoneme-levelfeaturesadaptivephonemepoolinggraphattentionnetworkrandomsubstitutionaugmentationaudioanti-spoofingcross-datasetgeneralization
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 claims that synthesized speech differs from genuine speech in the fine-grained acoustics of phonemes and their transitions, and that this discrepancy is a reliable detection signal. It introduces adaptive phoneme pooling, which averages frame-level audio features over the time span of each predicted phoneme, converting a whole utterance into an ordered sequence of phoneme vectors. A graph attention network then models the temporal dependencies among consecutive phonemes, while a random phoneme substitution augmentation supplies extra training diversity. Across four benchmarks, the resulting detector outperforms nine existing methods, with a 7.12% equal error rate on ASVspoof2021 DF and 16.07% on InTheWild, and retains strong performance under noise and compression. The paper's central assertion is that modern neural synthesizers can sound realistic but cannot faithfully reproduce the vocal-tract dynamics encoded in phoneme-level features.

What carries the argument

The central machinery is adaptive phoneme pooling. A multilingual phoneme recognition model, trained with CTC loss on roughly 375k Common Voice samples in nine languages (with a phoneme error rate around 0.4), predicts a phoneme label for every audio frame; consecutive frames sharing the same label are averaged into a single phoneme vector, producing a sample-specific phoneme-level feature sequence. That sequence is passed to a graph attention network in which each phoneme connects to its next N-1 neighbors (N=10), and attention coefficients weight how each phoneme depends on the ones following it. The augmentations and alignment loss (random phoneme substitution, CLIP-style loss) are secondary but help the model exploit the phoneme-level signal.

What would settle it

Run the detector with phoneme segments derived from ground-truth forced alignments (on a dataset where such alignments exist, e.g., LibriSpeech) instead of the recognizer's predicted labels. If the detection advantage mostly disappears with true alignments, the 'phoneme inconsistencies' are recognition artifacts, not acoustic properties of synthesizers.

Watch

Extended reading notes

Core claim

The paper's discovery is that deepfake speech exhibits systematic inconsistencies at the phoneme level: the individual representations of phonemes and the transitions between them differ from bonafide speech in a way that generalizes across synthesizers, languages, and unseen conditions. The authors demonstrate this by pooling frame-level features (from WavLM or Wav2Vec2) according to predicted phoneme boundaries, which yields phoneme-level features whose t-SNE projections separate real from fake samples more cleanly than frame-level features. Building on this observation, they construct a detector that feeds phoneme-level features through a graph attention network to capture temporal dependencies of phoneme transitions, and they introduce random phoneme substitution augmentation, which swaps phoneme segments between samples and labels them fake, plus a CLIP-style alignment loss that ties the detector's frame representations to the recognizer's phonetic representations. This model achieves the best reported results on all four tested datasets, including the InTheWild challenge set, where it reaches a 16.07% equal error rate.

Load-bearing premise

The method assumes the phoneme recognizer's segmentation is accurate enough that pooling over its predicted phoneme segments extracts genuine phonetic structure; a phoneme error rate near 0.4 leaves room for recognition failures, especially on synthetic audio, to masquerade as phoneme-level inconsistencies.

Editorial extensions

If this is right

  • Phoneme-based detection becomes annotation-free: the pre-trained recognizer supplies phoneme labels and timestamps automatically, so the method works on any language and on unseen deepfake generators.
  • The phoneme-level signal is cross-lingual: models trained on English, German, and Spanish transfer to French, Italian, Polish, Russian, Ukrainian, and the wild InTheWild recordings, suggesting a common acoustic shortfall in synthetic phoneme transitions.
  • The detector stays strong under background noise (20 dB SNR) and compression artifacts, indicating the phoneme inconsistency is not just a high-frequency artifact but part of the phoneme's structure.
  • The random phoneme substitution augmentation effectively manufactures fake training examples from real speech, which could be reused in other speech-forensics tasks.

Reading between the lines

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

  • The dependency on a phoneme recognizer with roughly 40% error rate creates an unexamined risk: if the recognizer fails systematically differently on synthetic audio, the detector might be exploiting recognition artifacts rather than phoneme physics; comparing performance with ground-truth forced alignments would settle this.
  • A natural next step would be an adversarial synthesizer trained to match real phoneme-level feature statistics; whether this detector resists such 'phoneme-aware' forgery is not addressed in the paper and is a concrete test of the claimed generalization.
  • Because the recognizer was trained on nine mostly European languages, performance on tonal languages (e.g., Mandarin Chinese, Thai) is untested; the paper reports no per-language results for Chinese in MLAAD even though the dataset includes it.
  • The graph attention over phoneme transitions is a lightweight phonetic model; extending it to a Transformer or convolutional decoder over phoneme sequences could capture longer-range dependencies than the N=10 lookahead and might improve robustness further.
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

5 major / 6 minor

Summary. The paper proposes a deepfake speech detector built on phoneme-level feature analysis. The authors pre-train a multilingual phoneme recognizer (Wav2Vec2/WavLM backbone) with CTC on Common Voice, freeze it, and use its frame-level phoneme predictions to perform adaptive average pooling over frame-level features extracted by the same backbone. The resulting phoneme-level features are fed into a graph attention network (GAT) with edges between consecutive phonemes, followed by an LSTM and a classification head. A random phoneme substitution augmentation (RPSA) mixes phoneme segments across samples and labels the result as fake, and a CLIP-style loss aligns the copied transformer's frame-level features with the frozen recognizer's features. Experiments on ASVspoof2019 LA, ASVspoof2021 DF, MLAAD, and InTheWild report state-of-the-art AUC/EER, including 7.12% EER on ASVspoof2021 DF and 16.07% EER on InTheWild, with ablations showing contributions from each component.

Significance. If the findings hold, the paper makes a useful contribution to audio deepfake detection: it proposes a novel mechanism (phoneme-level inconsistencies) and a strong practical system that generalizes across methods, languages, and datasets, outperforming several recent baselines. The ablations are informative and the robustness experiments (noise, compression) add practical credibility. However, the central conceptual claim—that deepfake speech exhibits intrinsic phoneme-level inconsistencies—is not fully established, because the phoneme labels come from a recognizer with a high PER (≈0.4) and the authors do not test whether the detected signal is an artifact of recognition errors on synthetic audio. The strongest contribution is empirical; the theoretical framing needs additional support.

major comments (5)
  1. [Pretraining Phoneme Recognition Model / Adaptive Phoneme Pooling] The paper reports a phoneme error rate of approximately 0.4 for the multilingual phoneme recognizer, yet all phoneme-level features in the detection pipeline are obtained by pooling frames according to these predicted labels. Since the recognizer was trained on Common Voice (natural speech), its errors on out-of-distribution synthetic audio may be systematically different, so the observed 'phoneme-level inconsistencies' could reflect recognition failure rather than intrinsic acoustic properties of deepfakes. The manuscript should report PER, confidence, or a segmentation-quality measure on bonafide versus fake test samples, and ideally compare adaptive pooling with oracle phoneme alignments (e.g., from available transcripts) on a subset to support the central claim.
  2. [Random Phoneme Substitution Augmentation] In the RPSA method (Section 'Random Phoneme Substitution Augmentation'), augmented samples are always labeled fake regardless of the original sample's label, meaning that real speech mixed with phonemes from another real speaker is presented as a deepfake. This may teach the model to detect the augmentation artifact (e.g., speaker-mismatch discontinuities) rather than genuine synthesis artifacts. The paper should analyze how much of the test performance is attributable to this augmentation signal, for example by evaluating the model on real fake samples while ablating RPSA, or by measuring whether the detector flags original real samples differently when RPSA is removed.
  3. [Experiment Setting / Table 1 / Supplementary Table 7] The evaluation protocol for ASVspoof2021 DF rebalances the test set by subsampling fake samples to match the number of bonafide samples, and all evaluation uses 3-second clips. These choices deviate from the official ASVspoof2021 evaluation protocol, and the rebalancing can materially change the EER and make the reported 7.12% not directly comparable with published results that use the full test set. The authors should justify the rebalancing and clipping, or report results under the official protocol in addition to their modified protocol.
  4. [Table 5, Ablation Study] The claim that adaptive phoneme pooling brings about a 16.7% EER improvement is based on comparing setting (f) with setting (d), but setting (d) removes GAT and RPSA as well as pooling. Because GAT and RPSA depend on pooling, the improvement cannot be isolated to the pooling operation. The authors should compare adaptive pooling against a non-adaptive alternative (e.g., fixed-length segmentation) while keeping GAT and RPSA active, to support the title's claim that phoneme-level discrepancies are the key.
  5. [Tables 1–5] No confidence intervals, standard deviations, or number of random seeds are reported. Given the observed variability in deepfake detection and the small margins between some baselines, the claim of 'significantly outperforming' all baselines requires at least a few repeated runs with reported mean and variance, especially for the main ASVspoof2021 DF and InTheWild results.
minor comments (6)
  1. [Pretraining Phoneme Recognition Model] The sentence says the model is trained on 9 languages but lists only 8 in parentheses (EN, DE, ES, FR, PL, RU, UK, ZH); the ninth language should be identified.
  2. [Graph Attention Module] The edge construction rule for N=10 can add duplicate edges to the same node when i is near T′ (e.g., i=T′−1 adds multiple edges to node T′). The notation should be refined to avoid self-loop or duplicate edges.
  3. [Loss Function] Equation (4) is written compactly; the role of the projection g and the indices i,k should be clarified to avoid ambiguity about which feature is the anchor and which are negatives.
  4. [Throughout] The paper contains several typos and inconsistent notation, e.g., 'Implementaion Details', 'T-SNE' versus 't-SNE', and the use of 'phoneme features' vs 'phoneme-level features' in places; a careful proofread is needed.
  5. [Experiment Setting] The description of the 'seen' and 'unseen' synthesizer split for ASVspoof2021 DF is brief; the mapping to the official dataset partitions (e.g., which attacks are 'seen') should be stated explicitly in the main text.
  6. [Supplementary, Training of Phoneme Recognition Model] The PER formula (Eq. 6) defines an error rate, but the text 'PER denotes the accuracy' is misleading; this should be corrected to 'error rate'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the detector is trained on labeled real/fake data and evaluated on held-out benchmark splits, with phoneme labels from an independently pretrained recognizer.

full rationale

The paper's central claim is that deepfake speech shows phoneme-level feature inconsistencies. The claimed derivation chain is: (1) pretrain a multilingual phoneme recognizer on Common Voice 6.1, independent of all deepfake evaluation sets; (2) use it to produce phoneme labels for adaptive pooling; (3) train a GAT-based detector on labeled train/validation subsets of benchmark datasets; (4) evaluate on held-out test subsets (ASVspoof2021 DF, MLAAD held-out languages, InTheWild). No parameter in the pipeline is fitted to the test labels, and no reported EER is a renamed training objective. The pooling operation is a deterministic averaging of consecutive frames with the same predicted phoneme, not a fitted constant. The t-SNE visualizations are descriptive illustrations and do not enter the training loss. The ablation study demonstrates that the pooling, GAT, CLIP loss, and RPSA each affect the held-out EER, which is evidence that the components are doing real work. The only notable concern - that the 0.4 phoneme error rate of the recognizer could cause different segmentation errors on out-of-distribution synthetic audio, so the phoneme-level inconsistencies might partly reflect recognition artifacts - is a correctness/generalization risk, not circularity: the recognizer was trained on a public corpus with no access to the deepfake test sets, and the detection labels are external. Because the central results are benchmark-prediction results rather than assumptions re-imported as conclusions, the derivation is self-contained and no circular step can be exhibited.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central claim depends on several hand-chosen hyperparameters and on domain assumptions about phonemes and synthesizer limitations. No new physical or mathematical entities are introduced.

free parameters (6)
  • substitution probability p = 0.2
    Set for the RPSA augmentation; controls how often phonemes are replaced during training and affects detection performance.
  • number of GAT edges N = 10
    Neighborhood size for graph attention connections between phonemes; chosen by hand.
  • CLIP loss weight = 0.5
    Coefficient applied to LCLIP and L'_cls in the final objective L = Lcls + 0.5 * (LCLIP + L'_cls).
  • early stopping patience = 3 epochs
    Training stops if AUC does not improve for three consecutive epochs.
  • learning rates = 5e-5 (Transformer), 1e-4 (others)
    AdamW learning rates for the copied Transformer encoder and the remaining learnable parameters.
  • audio clip length = 3 seconds
    Three-second clips are randomly selected during training; the middle three seconds are used for validation and testing.
assumptions (4)
  • domain assumption Phonemes are the fundamental building blocks of speech, and each phoneme corresponds to a unique vocal tract configuration.
    This motivates the entire phoneme-level feature design; stated in the Introduction.
  • domain assumption Current synthesizers cannot perfectly simulate the ingredients of words, transitions between words, and emotional style, leading to detectable phoneme-level differences.
    This is the core mechanistic assumption; stated in the Introduction and Phoneme-level Feature Analysis.
  • domain assumption The multilingual phoneme recognition model trained on Common Voice transfers to deepfake datasets without adaptation.
    The detector relies on the recognizer's phoneme labels for pooling; stated in Pretraining Phoneme Recognition Model.
  • standard math CTC alignment can map frame-level predictions to phoneme sequences.
    Used for training the phoneme recognizer; from Graves et al. 2006.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Phoneme-Level Feature Discrepancies: A Key to Detecting Sophisticated Speech Deepfakes." pith.science (2026). https://pith.science/paper/CTY6PAVJ

@misc{pith2026241212619,
  author       = {Pith},
  title        = {Pith review of: Phoneme-Level Feature Discrepancies: A Key to Detecting Sophisticated Speech Deepfakes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CTY6PAVJ}},
  note         = {Machine review of arXiv:2412.12619}
}
read the original abstract

Recent advancements in text-to-speech and speech conversion technologies have enabled the creation of highly convincing synthetic speech. While these innovations offer numerous practical benefits, they also cause significant security challenges when maliciously misused. Therefore, there is an urgent need to detect these synthetic speech signals. Phoneme features provide a powerful speech representation for deepfake detection. However, previous phoneme-based detection approaches typically focused on specific phonemes, overlooking temporal inconsistencies across the entire phoneme sequence. In this paper, we develop a new mechanism for detecting speech deepfakes by identifying the inconsistencies of phoneme-level speech features. We design an adaptive phoneme pooling technique that extracts sample-specific phoneme-level features from frame-level speech data. By applying this technique to features extracted by pre-trained audio models on previously unseen deepfake datasets, we demonstrate that deepfake samples often exhibit phoneme-level inconsistencies when compared to genuine speech. To further enhance detection accuracy, we propose a deepfake detector that uses a graph attention network to model the temporal dependencies of phoneme-level features. Additionally, we introduce a random phoneme substitution augmentation technique to increase feature diversity during training. Extensive experiments on four benchmark datasets demonstrate the superior performance of our method over existing state-of-the-art detection methods.

Figures

Figures reproduced from arXiv: 2412.12619 by the authors.

Figure 1
Figure 1. T-SNE cluster results. We first employ a pre [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Adaptive phoneme pooling process. Consecutive [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Training of multilingual phoneme recognition model and T-SNE cluster results of phoneme-level speech features. Af [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Overview of our deepfake detection model. Given the input feature, our model first uses a pre-trained phoneme [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references · 24 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  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]

    Baevski, A.; Zhou, Y.; Mohamed, A.; and Auli, M. 2020. wav2vec 2.0: A framework for self-supervised learning of speech representations. Advances in Neural Information Processing Systems, 33: 12449--12460

    work page 2020

  4. [4]

    Blue, L.; Warren, K.; Abdullah, H.; Gibson, C.; Vargas, L.; O'Dell, J.; Butler, K.; and Traynor, P. 2022. Who are you (i really wanna know)? detecting audio \ DeepFakes \ through vocal tract reconstruction. In 31st USENIX Security Symposium (USENIX Security 22), 2691--2708

    work page 2022

  5. [5]

    D.; Junior, A

    Casanova, E.; Weber, J.; Shulby, C. D.; Junior, A. C.; G \"o lge, E.; and Ponti, M. A. 2022. YourTTS : Towards Zero-Shot Multi-Speaker TTS and Zero-Shot Voice Conversion for Everyone . In Proceedings of the 39th International Conference on Machine Learning , 2709--2720. PMLR

    work page 2022

  6. [6]

    Chen, S.; Wang, C.; Chen, Z.; Wu, Y.; Liu, S.; Chen, Z.; Li, J.; Kanda, N.; Yoshioka, T.; Xiao, X.; Wu, J.; Zhou, L.; Ren, S.; Qian, Y.; Qian, Y.; Wu, J.; Zeng, M.; Yu, X.; and Wei, F. 2022. WavLM : Large - Scale Self - Supervised Pre - Training for Full Stack Speech Processing . IEEE Journal of Selected Topics in Signal Processing, 1505--1518

    work page 2022

  7. [7]

    A.; and Raza, A

    Dhamyal, H.; Ali, A.; Qazi, I. A.; and Raza, A. A. 2021. Using Self Attention DNNs to Discover Phonemic Features for Audio Deep Fake Detection . In 2021 IEEE Automatic Speech Recognition and Understanding Workshop ( ASRU ) , 1178--1184

    work page 2021

  8. [8]

    Graves, A.; Fernández, S.; Gomez, F.; and Schmidhuber, J. 2006. Connectionist temporal classification: labelling unsegmented sequence data with recurrent neural networks. In Proceedings of the 23rd international conference on Machine learning , ICML '06, 369--376. New York, NY, USA: Association for Computing Machinery. ISBN 978-1-59593-383-6

    work page 2006

Show all 33 references
  1. [9]

    Guan, W.; Li, Y.; Li, T.; Huang, H.; Wang, F.; Lin, J.; Huang, L.; Li, L.; and Hong, Q. 2024. MM-TTS: Multi-Modal Prompt Based Style Transfer for Expressive Text-to-Speech Synthesis. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, 18117--18125

    2024

  2. [10]

    Guzhov, A.; Raue, F.; Hees, J.; and Dengel, A. 2022. Audioclip: Extending Clip to Image , Text and Audio . In ICASSP 2022 - 2022 IEEE International Conference on Acoustics , Speech and Signal Processing ( ICASSP ) , 976--980

    2022

  3. [11]

    S.; Lee, B.-J.; Yu, H.-J.; and Evans, N

    Jung, J.-w.; Heo, H.-S.; Tak, H.; Shim, H.-j.; Chung, J. S.; Lee, B.-J.; Yu, H.-J.; and Evans, N. 2022. Aasist: Audio anti-spoofing using integrated spectro-temporal graph attention networks. In ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Pro...

    2022

  4. [12]

    Jung, J.-w.; Kim, S.-b.; Shim, H.-j.; Kim, J.-h.; and Yu, H.-J. 2020. Improved RawNet with Feature Map Scaling for Text-independent Speaker Verification using Raw Waveforms. Proc. Interspeech, 3583--3587

    2020

  5. [13]

    Kim, J.; Kong, J.; and Son, J. 2021. Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech . In Proceedings of the 38th International Conference on Machine Learning , 5530--5540. PMLR

    2021

  6. [14]

    F.; Bakhturina, E.; Desta, M.; Valle, R.; Yoon, S.; and Catanzaro, B

    Kim, S.; Shih, K.; Badlani, R.; Santos, J. F.; Bakhturina, E.; Desta, M.; Valle, R.; Yoon, S.; and Catanzaro, B. 2023. P- Flow : A Fast and Data-Efficient Zero-Shot TTS through Speech Prompting . Advances in Neural Information Processing Systems, 36: 74213--74228

    2023

  7. [15]

    Lavrentyeva, G.; Novoselov, S.; Tseren, A.; Volkova, M.; Gorlanov, A.; and Kozlov, A. 2019. STC antispoofing systems for the ASVspoof2019 challenge. arXiv preprint arXiv:1904.05576

    2019 arXiv

  8. [16]

    Liu, X.; Wang, X.; Sahidullah, M.; Patino, J.; Delgado, H.; Kinnunen, T.; Todisco, M.; Yamagishi, J.; Evans, N.; Nautsch, A.; and Lee, K. A. 2023. ASVspoof 2021: Towards Spoofed and Deepfake Speech Detection in the Wild . IEEE/ACM Transactions on Audio, Speech, and Language Pr...

    2023

  9. [17]

    Loshchilov, I.; and Hutter, F. 2019. Decoupled Weight Decay Regularization . arXiv:1711.05101

    2019 arXiv

  10. [18]

    u ller, N.; Czempin, P.; Diekmann, F.; Froghyar, A.; and B \

    M \"u ller, N.; Czempin, P.; Diekmann, F.; Froghyar, A.; and B \"o ttinger, K. 2022. Does Audio Deepfake Detection Generalize ? In Interspeech 2022, 2783--2787. ISCA

    2022

  11. [19]

    u ller, N. M.; Kawa, P.; Choong, W. H.; Casanova, E.; G \

    M \"u ller, N. M.; Kawa, P.; Choong, W. H.; Casanova, E.; G \"o lge, E.; M \"u ller, T.; Syga, P.; Sperl, P.; and B \"o ttinger, K. 2024. MLAAD : The Multi-Language Audio Anti-Spoofing Dataset . arXiv:2401.09512

    2024 arXiv

  12. [20]

    Oord, A. v. d.; Dieleman, S.; Zen, H.; Simonyan, K.; Vinyals, O.; Graves, A.; Kalchbrenner, N.; Senior, A.; and Kavukcuoglu, K. 2016. WaveNet: A generative model for raw audio. arXiv preprint arXiv:1609.03499

    2016 arXiv

  13. [21]

    Qi, T.; Zheng, W.; Lu, C.; Zong, Y.; and Lian, H. 2024. PAVITS : Exploring Prosody-Aware VITS for End-to-End Emotional Voice Conversion . In ICASSP 2024 - 2024 IEEE International Conference on Acoustics , Speech and Signal Processing ( ICASSP ) , 12697--12701

    2024

  14. [22]

    Simonyan, K.; and Zisserman, A. 2015. Very Deep Convolutional Networks for Large-Scale Image Recognition . arXiv:1409.1556

    2015 arXiv

  15. [23]

    Snyder, D.; Chen, G.; and Povey, D. 2015. MUSAN : A Music , Speech , and Noise Corpus . arXiv:1510.08484

    2015 arXiv

  16. [24]

    Sun, C.; Jia, S.; Hou, S.; and Lyu, S. 2023. AI-Synthesized Voice Detection Using Neural Vocoder Artifacts . In Proceedings of the IEEE / CVF Conference on Computer Vision and Pattern Recognition , 904--912

    2023

  17. [25]

    Tan, X.; Chen, J.; Liu, H.; Cong, J.; Zhang, C.; Liu, Y.; Wang, X.; Leng, Y.; Yi, Y.; He, L.; Zhao, S.; Qin, T.; Soong, F.; and Liu, T.-Y. 2024. NaturalSpeech : End-to-End Text-to-Speech Synthesis With Human-Level Quality . IEEE Transactions on Pattern Analysis and Machine Int...

    2024

  18. [26]

    Van der Maaten, L.; and Hinton, G. 2008. Visualizing data using t- SNE . Journal of machine learning research, 9(11)

    2008

  19. [27]

    Veli c kovi \'c , P.; Cucurull, G.; Casanova, A.; Romero, A.; Lio, P.; and Bengio, Y. 2017. Graph attention networks. arXiv preprint arXiv:1710.10903

    2017 arXiv

  20. [28]

    Y.; and Zhang, X

    Wang, C.; He, J.; Yi, J.; Tao, J.; Zhang, C. Y.; and Zhang, X. 2024. Multi- Scale Permutation Entropy for Audio Deepfake Detection . In ICASSP 2024 - 2024 IEEE International Conference on Acoustics , Speech and Signal Processing ( ICASSP ) , 1406--1410

    2024

  21. [29]

    A.; Juvela, L.; Alku, P.; Peng, Y.-H.; Hwang, H.-T.; Tsao, Y.; Wang, H.-M.; Maguer, S

    Wang, X.; Yamagishi, J.; Todisco, M.; Delgado, H.; Nautsch, A.; Evans, N.; Sahidullah, M.; Vestman, V.; Kinnunen, T.; Lee, K. A.; Juvela, L.; Alku, P.; Peng, Y.-H.; Hwang, H.-T.; Tsao, Y.; Wang, H.-M.; Maguer, S. L.; Becker, M.; Henderson, F.; Clark, R.; Zhang, Y.; Wang, Q.; J...

    2020

  22. [30]

    M.; Gulzar, R.; and Amerini, I

    Wani, T. M.; Gulzar, R.; and Amerini, I. 2024. ABC-CapsNet : Attention Based Cascaded Capsule Network for Audio Deepfake Detection . In Proceedings of the IEEE / CVF Conference on Computer Vision and Pattern Recognition , 2464--2472

    2024

  23. [31]

    Wu, H.-H.; Seetharaman, P.; Kumar, K.; and Bello, J. P. 2022. Wav2CLIP : Learning Robust Audio Representations from Clip . In ICASSP 2022 - 2022 IEEE International Conference on Acoustics , Speech and Signal Processing ( ICASSP ) , 4563--4567

    2022

  24. [32]

    Xie, Y.; Cheng, H.; Wang, Y.; and Ye, L. 2024. Domain Generalization via Aggregation and Separation for Audio Deepfake Detection . IEEE Transactions on Information Forensics and Security, 19: 344--358

    2024

  25. [33]

    Y.; Zeng, S.; and Tao, J

    Zhang, X.; Yi, J.; Wang, C.; Zhang, C. Y.; Zeng, S.; and Tao, J. 2024. What to Remember : Self-Adaptive Continual Learning for Audio Deepfake Detection . Proceedings of the AAAI Conference on Artificial Intelligence, 38(17): 19569--19577

    2024

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.