Pith. sign in

REVIEW 4 major objections 6 minor 4 cited by

Investigating Prosodic Signatures via Speech Pre-Trained Models for Audio Deepfake Source Attribution

T0 review · 4 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 Fusing Whisper and x-vector embeddings with a Rényi-divergence loss attributes audio deepfakes to their generating system, reaching 98.91% accuracy on ASVspoof 2019 and 99.01% accuracy on CFAD.

desk verdict A useful PTM comparison for audio deepfake source attribution, but the ASVspoof protocol leaks speaker identity and the SOTA claim does not survive a speaker-disjoint evaluation. read the letter →

arxiv 2412.17796 v1 pith:KYDT2Q4D submitted 2024-12-23 eess.AS cs.SD

Orchid Chetia Phukan , Drishti Singh , Swarup Ranjan Behera , Arun Balaji Buduru , Rajesh Sharma This is my paper · ORCID
classification eess.AScs.SD
keywords audiodeepfakedetectionsourceattributionprosodicsignaturespre-trainedspeechmodelsx-vectorRényidivergencefeaturefusionASVspoof2019
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

Audio deepfakes carry prosodic fingerprints—pitch, tone, rhythm, intonation—that differ from one text-to-speech or voice-conversion generator to the next. The paper asks which speech pretrained models best expose those fingerprints for source attribution, and finds that x-vector, a compact speaker-recognition embedding, outperforms larger general-purpose models such as Whisper and XLS-R. It then proposes FINDER, a fusion method that uses Rényi divergence to align two pretrained representation spaces while training a classifier, and reports that Whisper plus x-vector fused this way gives the best closed-set attribution: 98.91% accuracy with 0.26% equal error rate on ASVspoof 2019 and 99.01% accuracy with 1.07% equal error rate on CFAD. If the result holds, forensic attribution of a deepfake to its generating tool can be done by freezing pretrained speech embeddings and training only a small head.

What carries the argument

Two mechanisms carry the argument. First, x-vector is a time-delay neural network embedding trained for speaker recognition; the paper treats its learned speaker-prosody sensitivity as a transferable fingerprint extractor for generator identity. Second, FINDER fuses two frozen pretrained-model representation spaces by projecting each through convolutional blocks, then adding a Rényi-divergence loss $L_{\mathrm{RD}} = \frac{1}{\alpha-1}\log \sum_i (e_{a,i}+\epsilon)^\alpha(e_{b,i}+\epsilon)^{1-\alpha}$ to the cross-entropy loss. The divergence term pulls the two representations toward each other while the classifier separates generator classes, with hyperparameters $\alpha=2$, $\epsilon=0.1$, and $\lambda=0.4$ weighting the two losses. This alignment is what lets the complementary ASR-style information of Whisper and speaker-style information of x-vector combine more effectively than concatenation.

What would settle it

Re-run the exact FINDER (Whisper + x-vector) pipeline on ASVspoof 2019 and CFAD with speaker-disjoint and recording-disjoint folds in place of the random split, and compare accuracy and equal error rate to 98.91%/0.26% and 99.01%/1.07%; a large rise in EER would show the headline result depends on the optimistic split rather than on source prosody.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that prosodic source signatures are best captured by a speaker-recognition pretrained model, x-vector, and that fusing those embeddings with Whisper's via the proposed FINDER loss is the strongest configuration. In controlled comparisons across six pretrained models and two downstream heads, x-vector needed only about 4.2 million parameters yet beat larger monolingual and multilingual models on both benchmark datasets, which the authors attribute to speaker-recognition pretraining. Replacing simple concatenation with their Rényi-divergence fusion consistently improved results for nearly every pretrained-model pair. The best pair, Whisper and x-vector, reaches 98.91% accuracy with 0.26% equal error rate on ASVspoof 2019 and 99.01% accuracy with 1.07% equal error rate on CFAD, ahead of the reimplemented MiO and AASIST baselines. Those numbers are closed-set: the system can only attribute audio to generator classes seen in training.

Load-bearing premise

The load-bearing assumption is that the random five-fold split of the combined ASVspoof 2019 partitions, without speaker-disjoint or recording-disjoint folds, is a valid measure of source attribution; if that split is optimistic, the reported numbers are inflated.

Editorial extensions

If this is right

  • Closed-set attribution of deepfake audio can be built from frozen pretrained embeddings plus a small trainable classifier, without training a large audio model from scratch.
  • Speaker-recognition pretraining appears more useful for prosodic source fingerprints than much larger self-supervised speech models; model size is not the controlling factor.
  • Fusing an ASR-oriented model with a speaker-oriented model through divergence alignment beats concatenation of the same embeddings.
  • The paper gives the first audio-deepfake source-attribution numbers on CFAD and a new baseline for ASVspoof 2019 source attribution.
  • The system does not generalize to generators absent from the training set; open-vocabulary attribution remains unsolved.

Reading between the lines

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

  • The paper does not enforce speaker-disjoint or recording-disjoint folds on ASVspoof 2019, so the reported 98.91% accuracy and 0.26% EER could be lower under a split that prevents the same speaker or recording from appearing in both training and test folds.
  • A testable extension, not in the paper, is adding explicit prosody features such as pitch, energy, or rhythm to the FINDER alignment; this would show whether prosody itself, rather than channel or recording artifacts, drives the result.
  • The Rényi-divergence fusion loss is model-agnostic and could be evaluated as a generic multi-encoder fusion loss for other speech tasks such as emotion recognition or speaker verification.
  • The paper's own limitations section notes the closed-vocabulary setting; holding out entire generator classes would test whether the approach can attribute audio to TTS or VC systems never seen in training.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper investigates speech pre-trained models (PTMs) for audio deepfake source attribution (ADSD), focusing on prosodic signatures of generative sources. It compares six PTMs (wav2vec2, wav2vec2-emo, XLS-R, Whisper, WavLM, x-vector) with FCN and CNN downstream networks on ASVspoof 2019 and CFAD. It further proposes FINDER, a fusion method that adds a Rényi-divergence loss between projected PTM representations to the cross-entropy loss. The main empirical claims are: (i) x-vector is the best individual PTM, attributed to speaker-recognition pre-training capturing prosodic characteristics; (ii) fusion of Whisper and x-vector representations through FINDER achieves the topmost performance, with 98.91% accuracy / 0.26% EER on ASV and 99.01% accuracy / 1.07% EER on CFAD, described as state-of-the-art (SOTA).

Significance. The paper addresses a relevant and timely problem, audio deepfake source attribution, and provides a broad comparison of established PTMs. Its proposed fusion loss is simple and the experimental matrix is internally consistent. The strongest contribution is the empirical evidence that fusion of complementary PTM representations improves attribution over individual PTMs, and that FINDER consistently outperforms concatenation baselines under the same evaluation setup. The paper also ships useful t-SNE visualizations and promises code release. However, the central quantitative claims and the prosodic interpretation rest on an evaluation protocol that is not speaker-disjoint or recording-disjoint, which calls into question the absolute accuracy/EER numbers and the SOTA label. If the protocol issue is fixed and the results remain qualitatively similar, the paper would be a worthwhile contribution; in its current form, the main claims are not trustworthy as stated.

major comments (4)
  1. [Section 5, Benchmark Datasets; Section 5, Comparison to Previous Works; Table 3] The ASV evaluation protocol is the load-bearing element of the paper's SOTA claim. The text states: 'We combine the train, validation and testing splits for ASV' and 'We followed 5-fold cross validation for ASV,' with no mention of speaker-disjoint or recording-disjoint folds. ASVspoof 2019 LA was constructed with distinct speakers across its original partitions, and the spoofed A-class samples are generated from shared bonafide source utterances. Randomly splitting the pooled utterances into 5 folds therefore makes it very likely that the same speaker, and even the same utterance content, appears in both training and test folds. The downstream CNN, particularly on x-vector speaker embeddings, can exploit this identity leakage as a shortcut. This inflates the reported 98.91% accuracy / 0.26% EER and confounds the attribution interpretation. The authors must either enforce speaker/recording-disjoint folds (e.g., by grouping utterances by source recording and speaker) and report results under that protocol, or explicitly justify why random folds constitute a valid attribution evaluation.
  2. [Section 5, Comparison to Previous Works; Table 3; Abstract; Conclusion] The paper states 'as we have considered all the source classes across train, validation, and test split for ASV, so we can't directly compare our results to previous works,' yet the abstract, conclusion, and Table 3 still claim SOTA performance. The reimplemented baselines (MiO and AASIST) in Table 3 are evaluated under the same non-standard protocol, so the relative comparison is internally consistent; however, the claim of 'attaining SOTA performance' against the published literature is unsupported. Please either temper the claim to 'SOTA under our evaluation protocol' or provide a protocol-matched comparison using published numbers where the original train/dev/test split is respected.
  3. [Section 5, Experimental Results; Table 1; Section 6, Conclusion] The interpretation that x-vector outperforms other PTMs because its speaker-recognition pre-training captures prosodic source signatures is confounded by the protocol leakage. Under a random split of pooled utterances, x-vector embeddings give the most direct access to speaker identity, which is an even simpler cue than prosody. To support the prosody-based explanation, the authors need to control for speaker/recording overlap, for example by evaluating on speaker-disjoint folds or by ablating speaker-related dimensions of the x-vector embedding. Without such a control, the observed x-vector advantage cannot be attributed to prosodic signature capture.
  4. [Section 5, Training Details; Table 2] The hyperparameters α, ε, and λ were chosen 'through some preliminary exploration' and the best fusion pair (Whisper + x-vector) is the selected maximum over the fourteen fusion pairs in Table 2. The reported 98.91% and 99.01% numbers are therefore a selected maximum over hyperparameters and fusion pairs, not an independent out-of-sample prediction. The authors should describe the model selection procedure (e.g., a separate validation split) or at least report the variance across the 5 folds and disclose that the best pair is chosen on the evaluation set.
minor comments (6)
  1. [Section 4, FINDER equation] The Rényi divergence loss adds a small constant ε inside both (ea,i+ε) and (eb,i+ε) before the log and power operations; please clarify whether this shift is intended to bias the divergence estimate or solely to ensure numerical stability, and state how ε interacts with the choice of α.
  2. [Section 5, Benchmark Datasets; Table 2 caption] The caption of Table 2 says 'All the scores are average of 5-folds,' but the text says CFAD uses the 'official split given for CFAD.' Please clarify whether CFAD results are also averaged over folds or are single-run numbers from the official split.
  3. [Appendix A.1, last paragraph] The sentence 'We extract representations from the last hidden state of the PTMs by average pooling. We' is truncated; it should be completed or removed.
  4. [Appendix A.3, System Configurations] There is an unmatched opening parenthesis in 'max pooling (pool size 2.' and the following sentence also has a typo ('succeeded by followed by'). Please fix these punctuation errors.
  5. [Section 5, Comparison to Previous Works] The phrase 'theFINDER' appears without a space at the end of the comparison paragraph; also the capitalization 'FINDER (FusIon through Re Nyi DivERgence)' is unconventional and should be normalized.
  6. [Section 5, Experimental Results] The EER computation is described only as 'the average scores of one-vs-all.' Please specify how the one-vs-all EER is aggregated (e.g., macro-average over classes) and whether the EER is computed on scores or on probabilities, since this affects comparability with other work.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's claims are empirical measurements; the x-vector/prosody link is a post-hoc interpretation, and FINDER is a proposed loss objective rather than a derived prediction.

full rationale

The paper contains no derivation-level circularity. Its central results are measured accuracies and EERs from downstream models trained on frozen pre-trained representations; the reported ASV and CFAD numbers (e.g., 98.91% accuracy / 0.26% EER for FINDER with Whisper+x-vector) are experimental outcomes, not quantities fitted and then renamed as predictions. The claim that x-vector best captures prosodic signatures is explicitly offered as an interpretation ('This higher performance can be due to its speaker recognition pre-training that enables it for capturing unique prosodic characteristics'), not as a conclusion forced by a definition that presupposes it. FINDER's total loss L = λLCE + (1−λ)LRD is an optimization objective; no self-definitional equivalence between the loss and the reported accuracy is present. Self-citations to Chetia Phukan et al. 2023 and 2024 are used as background evidence for prior empirical behavior of x-vector in prosodic tasks and for PTM fusion in audio deepfake detection, and MiO is reimplemented as a baseline; these citations are not invoked as a uniqueness theorem or as the sole load-bearing justification for the paper's central claim. Concerns about the non-speaker-disjoint 5-fold split on ASV, the pooling of original train/dev/eval partitions, and hyperparameters selected from preliminary exploration are evaluation-validity and selection-bias concerns rather than circular reasoning; they affect the strength of the empirical claims but do not make the derivation equivalent to its inputs.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

No new physical or conceptual entities are introduced. The method FINDER is a loss function, not an entity. The central claim depends on several fitted hyperparameters and on assumptions about the validity of the evaluation protocol and the distributional interpretation of embeddings.

free parameters (4)
  • Rényi divergence order α = 2
    Fixed after preliminary exploration; controls the order of the divergence in FINDER (Section 5, Training Details).
  • loss weight λ = 0.4
    Chosen by preliminary exploration; balances cross-entropy and Rényi divergence losses in L = λL_CE + (1-λ)L_RD.
  • numerical stability constant ε = 0.1
    Set by preliminary exploration; added inside the Rényi divergence computation for numerical stability.
  • projected embedding dimension D = 120
    Features are linearly projected to 120 dimensions for dimensional alignment and computational constraints (Section 4).
assumptions (3)
  • domain assumption Rényi divergence is a valid regularizer when applied to raw pre-trained model feature embeddings as if they were probability distributions.
    Section 4 applies the Rényi divergence formula directly to projected features with no normalization or proof that they lie on a probability simplex; the theoretical guarantees for distributions are assumed to transfer.
  • ad hoc to paper Random 5-fold splitting of the combined ASVspoof train/dev/eval partitions yields a valid attribution evaluation.
    Section 5 abandons the original ASV protocol and does not enforce speaker- or recording-disjoint folds, so accuracy may be inflated by leakage between training and test folds.
  • domain assumption Prosodic characteristics are unique fingerprints of each generative source and dominate attribution performance.
    Section 1 asserts that prosodic signatures are one of the major fingerprints of generative sources, but no direct measurement of prosody is provided; the claim that better prosody capture explains ADSD accuracy is not directly tested.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Investigating Prosodic Signatures via Speech Pre-Trained Models for Audio Deepfake Source Attribution." pith.science (2026). https://pith.science/paper/KYDT2Q4D

@misc{pith2026241217796,
  author       = {Pith},
  title        = {Pith review of: Investigating Prosodic Signatures via Speech Pre-Trained Models for Audio Deepfake Source Attribution},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KYDT2Q4D}},
  note         = {Machine review of arXiv:2412.17796}
}
read the original abstract

In this work, we investigate various state-of-the-art (SOTA) speech pre-trained models (PTMs) for their capability to capture prosodic signatures of the generative sources for audio deepfake source attribution (ADSD). These prosodic characteristics can be considered one of major signatures for ADSD, which is unique to each source. So better is the PTM at capturing prosodic signs better the ADSD performance. We consider various SOTA PTMs that have shown top performance in different prosodic tasks for our experiments on benchmark datasets, ASVSpoof 2019 and CFAD. x-vector (speaker recognition PTM) attains the highest performance in comparison to all the PTMs considered despite consisting lowest model parameters. This higher performance can be due to its speaker recognition pre-training that enables it for capturing unique prosodic characteristics of the sources in a better way. Further, motivated from tasks such as audio deepfake detection and speech recognition, where fusion of PTMs representations lead to improved performance, we explore the same and propose FINDER for effective fusion of such representations. With fusion of Whisper and x-vector representations through FINDER, we achieved the topmost performance in comparison to all the individual PTMs as well as baseline fusion techniques and attaining SOTA performance.

Figures

Figures reproduced from arXiv: 2412.17796 by the authors.

Figure 1
Figure 1. Proposed Framework FINDER: RD and FCN stand for renyi divergence and fully connected network, respectively; L, LCE, and LRD represent the total loss, cross-entropy loss, and renyi divergence loss, respec￾tively. probability distributions (Van Erven and Harremos, 2014). Here, we frame RD as a loss function that calculates the divergence between the feature repre￾sentations of two different PTMs. We aim to reduce the … view at source ↗
Figure 2
Figure 2. Representation Space Visualization of PTMs for CFAD [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Representation Space Visualization of PTMs for ASV [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 4 Pith papers

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

  1. Open-Set Source Tracing of Audio Deepfake Systems

    eess.AS 2025-07 conditional novelty 5.0 of 10

    Softmax energy, a modified out-of-distribution score, improves open-set source tracing of audio deepfake systems, achieving a 31% relative FPR95 reduction and best FPR95 of 8.3% with augmentation.

  2. Attention Is Not Always the Answer: Optimizing Voice Activity Detection with Simple Feature Fusion

    cs.SD 2025-06 conditional novelty 5.0 of 10

    FusionVAD shows that simple addition or concatenation of MFCC and pre-trained model features outperforms cross-attention fusion for voice activity detection, with the best model beating Pyannote by 2.04 average DER.

  3. Source Tracing of Synthetic Speech Systems Through Paralinguistic Pre-Trained Representations

    eess.AS 2025-06 conditional novelty 5.0 of 10

    Paralinguistic speech representations, especially TRILLsson, are the most effective single features for tracing synthetic speech to its source generator, and the TRIO fusion with x-vector reports new accuracy highs.

  4. Towards Explainable Spoofed Speech Attribution and Detection:a Probabilistic Approach for Characterizing Speech Synthesizer Components

    eess.AS 2025-02 conditional novelty 5.0 of 10

    Probabilistic attribute embeddings derived from countermeasure embeddings match raw embedding performance on spoofed speech detection and attack attribution while providing component-level explanations.

Reference graph

Works this paper leans on

41 extracted references · 17 canonical work pages · cited by 4 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]

    Fake Biden robocall tells voters to skip New Hampshire primary election - BBC News

    2024. Fake Biden robocall tells voters to skip New Hampshire primary election - BBC News . https://www.bbc.com/news/world-us-canada-68064247. Last accessed: 2024-03-05

    work page 2024

  4. [4]

    Altalahin, S

    I. Altalahin, S. AlZu'bi, A. Alqudah, et al. 2023. Unmasking the truth: A deep learning approach to detecting deepfake audio through mfcc features. In Proc. of 2023 International Conference on Information Technology (ICIT), pages 511--518

    work page 2023

  5. [5]

    A Arunkumar, Vrunda Nileshkumar Sukhadia , and Srinivasan Umesh. 2022. https://doi.org/10.21437/Interspeech.2022-11376 Investigation of ensemble features of self-supervised pretrained models for automatic speech recognition . In Interspeech 2022, pages 5145--5149

    doi:10.21437/interspeech.2022-11376 2022

  6. [6]

    Arun Babu, Changhan Wang, Andros Tjandra, Kushal Lakhotia, Qiantong Xu, Naman Goyal, Kritika Singh, Patrick von Platen , Yatharth Saraf, Juan Pino, Alexei Baevski, Alexis Conneau, and Michael Auli. 2022. https://doi.org/10.21437/Interspeech.2022-143 Xls-r: Self-supervised cross-lingual speech representation learning at scale . In Interspeech 2022, pages 2...

    doi:10.21437/interspeech.2022-143 2022

  7. [7]

    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

  8. [8]

    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, 16(6):1505--1518

    2022

Show all 41 references
  1. [9]

    Orchid Chetia Phukan , Arun Balaji Buduru , and Rajesh Sharma. 2023. https://doi.org/10.21437/Interspeech.2023-2561 Transforming the embeddings: A lightweight technique for speech emotion recognition tasks . In INTERSPEECH 2023, pages 1903--1907

    2023 doi

  2. [10]

    Orchid Chetia Phukan, Gautam Kashyap, Arun Balaji Buduru, and Rajesh Sharma. 2024. https://doi.org/10.18653/v1/2024.findings-naacl.160 Heterogeneity over homogeneity: Investigating multilingual speech pre-trained models for detecting audio deepfake . In Findings of the Associa...

    2024 doi

  3. [11]

    Junlong Deng, Yanzhen Ren, Tong Zhang, Hongcheng Zhu, and Zongkun Sun. 2024. Vfd-net: Vocoder fingerprints detection for fake audio. In ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 12151--12155. IEEE

    2024

  4. [12]

    José Vicente Egas-López, Gábor Kiss, Dávid Sztahó, and Gábor Gosztolya. 2022. https://doi.org/10.1109/ICASSP43922.2022.9746068 Automatic assessment of the degree of clinical depression from speech using x-vectors . In ICASSP 2022 - 2022 IEEE International Conference on Acousti...

    2022

  5. [13]

    Tiantian Feng and Shrikanth Narayanan. 2023. Peft-ser: On the use of parameter efficient transfer learning approaches for speech emotion recognition using pre-trained speech models. In 2023 11th International Conference on Affective Computing and Intelligent Interaction (ACII)...

    2023

  6. [14]

    Takahiro Fukumori, Taito Ishida, and Yoichi Yamashita. 2023. Investigating the effectiveness of speaker embeddings for shout intensity prediction. In 2023 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC), pages 1838--1842. IEEE

    2023

  7. [15]

    Hamza, A

    A. Hamza, A. R. R. Javed, F. Iqbal, et al. 2022. Deepfake audio detection via mfcc features using machine learning. IEEE Access, 10:134018--134028

    2022

  8. [16]

    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

  9. [17]

    H. H. Kilinc and F. Kaledibi. 2023. Audio deepfake detection by using machine and deep learning. In Proc. of 2023 10th International Conference on Wireless Networks and Mobile Communications (WINCOM), pages 1--5

    2023

  10. [18]

    Kinnunen, M

    T. Kinnunen, M. Sahidullah, H. Delgado, N. Evans, M. Todisco, et al. 2017. The asvspoof 2017 challenge: Assessing the limits of replay spoofing attack detection. In Proc. of INTERSPEECH

    2017

  11. [19]

    Nicholas Klein, Tianxiang Chen, Hemlata Tak, Ricardo Casal, and Elie Khoury. 2024 a . https://doi.org/10.21437/Interspeech.2024-1283 Source tracing of audio deepfake systems . In Interspeech 2024, pages 1100--1104

    2024 doi

  12. [20]

    Nicholas Klein, Tianxiang Chen, Hemlata Tak, Ricardo Casal, and Elie Khoury. 2024 b . https://api.semanticscholar.org/CorpusID:271097599 Source tracing of audio deepfake systems . ArXiv, abs/2407.08016

    2024 arXiv

  13. [21]

    X. Liu, X. Wang, M. Sahidullah, et al. 2023. ASVspoof 2021 : Towards spoofed and deepfake speech detection in the wild. IEEE/ACM Transactions on Audio, Speech, and Language Processing

    2023

  14. [22]

    Haoxin Ma, Jiangyan Yi, Chenglong Wang, Xinrui Yan, Jianhua Tao, Tao Wang, Shiming Wang, and Ruibo Fu. 2024. Cfad: A chinese dataset for fake audio detection. Speech Communication, 164:103122

    2024

  15. [23]

    Nicolas Michael M \"u ller, Franziska Dieckmann, and Jennifer Williams. 2022. https://api.semanticscholar.org/CorpusID:247778506 Attacker attribution of audio deepfakes . ArXiv, abs/2203.15563

    2022 arXiv

  16. [24]

    Leonardo Pepino, Pablo Riera, and Luciana Ferrer. 2021. https://doi.org/10.21437/Interspeech.2021-703 Emotion recognition from speech using wav2vec 2.0 embeddings . In Interspeech 2021, pages 3400--3404

    2021 doi

  17. [25]

    Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, and Ilya Sutskever. 2023. Robust speech recognition via large-scale weak supervision. In International conference on machine learning, pages 28492--28518. PMLR

    2023

  18. [26]

    O. A. Shaaban, R. Yildirim, and A. A. Alguttar. 2023. Audio deepfake approaches. IEEE Access, 11:132652--132682

    2023

  19. [27]

    Jiatong Shi, Dan Berrebbi, William Chen, En-Pei Hu, Wei-Ping Huang, Ho-Lam Chung, Xuankai Chang, Shang-Wen Li, Abdelrahman Mohamed, Hung yi Lee, and Shinji Watanabe. 2023. https://doi.org/10.21437/Interspeech.2023-1316 Ml-superb: Multilingual speech universal performance bench...

    2023 doi

  20. [28]

    David Snyder, Daniel Garcia-Romero, Gregory Sell, Daniel Povey, and Sanjeev Khudanpur. 2018. https://doi.org/10.1109/ICASSP.2018.8461375 X-vectors: Robust dnn embeddings for speaker recognition . In 2018 IEEE International Conference on Acoustics, Speech and Signal Processing ...

    2018

  21. [29]

    Catherine Stupp. 2019. Fraudsters used ai to mimic ceo’s voice in unusual cybercrime case. The Wall Street Journal, 30(08)

    2019

  22. [30]

    Todisco, X

    M. Todisco, X. Wang, V. Vestman, Md. Sahidullah, and K. Lee. 2019. ASVspoof 2019 : Future horizons in spoofed and fake audio detection. In Proc. of INTERSPEECH

    2019

  23. [31]

    Tim Van Erven and Peter Harremos. 2014. R \'e nyi divergence and kullback-leibler divergence. IEEE Transactions on Information Theory, 60(7):3797--3820

    2014

  24. [32]

    Xin Wang, Junichi Yamagishi, Massimiliano Todisco, H \'e ctor Delgado, Andreas Nautsch, Nicholas Evans, Md Sahidullah, Ville Vestman, Tomi Kinnunen, Kong Aik Lee, et al. 2020. Asvspoof 2019: A large-scale public database of synthesized, converted and replayed speech. Computer ...

    2020

  25. [33]

    Z. Wu, T. Kinnunen, N. Evans, J. Yamagishi, C. Hanilçi, et al. 2015. ASVspoof 2015 : The first automatic speaker verification spoofing and countermeasures challenge. In Proc. of INTERSPEECH

    2015

  26. [34]

    Yamagishi, X

    J. Yamagishi, X. Wang, M. Todisco, M. Sahidullah, J. Patino, A. Nautsch, X. Liu, K. A. Lee, T. Kinnunen, and N. Evans. 2021. ASVspoof 2021 : Accelerating progress in spoofed and deepfake speech detection. In Proc. of INTERSPEECH

    2021

  27. [35]

    Xinrui Yan, Jiangyan Yi, Jianhua Tao, Chenglong Wang, Haoxin Ma, Zhengkun Tian, and Ruibo Fu. 2022 a . https://api.semanticscholar.org/CorpusID:251741101 System fingerprints detection for deepfake audio: An initial dataset and investigation . ArXiv, abs/2208.10489

    2022 arXiv

  28. [36]

    Xinrui Yan, Jiangyan Yi, Jianhua Tao, Chenglong Wang, Haoxin Ma, Tao Wang, Shiming Wang, and Ruibo Fu. 2022 b . An initial investigation for detecting vocoder fingerprints of fake audio. In Proc. of the 1st International Workshop on Deepfake Detection for Audio Multimedia

    2022

  29. [37]

    J. Yi, R. Fu, J. Tao, et al. 2022. ADD 2022 : The first audio deep synthesis detection challenge. In Proc. of ICASSP, pages 9216--9220

    2022

  30. [38]

    J. Yi, J. Tao, R. Fu, et al. 2023. ADD 2023 : The second audio deepfake detection challenge. arXiv preprint arXiv:2305.13774

    2023 arXiv

  31. [39]

    Salah Zaiem, Youcef Kemiche, Titouan Parcollet, Slim Essid, and Mirco Ravanelli. 2023. https://doi.org/10.21437/Interspeech.2023-1087 Speech self-supervised representation benchmarking: Are we doing it right? In INTERSPEECH 2023, pages 2873--2877

    2023 doi

  32. [40]

    Chu Yuan Zhang, Jiangyan Yi, Jianhua Tao, Chenglong Wang, and Xinrui Yan. 2023. https://api.semanticscholar.org/CorpusID:261705832 Distinguishing neural speech synthesis models through fingerprints in speech waveforms . ArXiv, abs/2309.06780

    2023 arXiv

  33. [41]

    Tinglong Zhu, Xingming Wang, Xiaoyi Qin, and Ming Li. 2022. Source tracing: Detecting voice spoofing. In Proc. Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC)

    2022

Pith tools

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