Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

Multilingual Source Tracing of Speech Deepfakes: A First Benchmark

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper establishes the first benchmark for tracing which text-to-speech model generated a deepfake speech sample across languages, and finds that a classic signal-processing front-end (LFCC with ECAPA-TDNN) generalizes across languages

desk verdict Useful first benchmark for multilingual source tracing, but the headline DSP-vs-SSL conclusion is undermined by a front-end/backend confound. read the letter →

arxiv 2508.04143 v1 pith:YM2TOJ4I submitted 2025-08-06 eess.AS cs.CLcs.SD

Xi Xuan , Yang Xiao , Rohan Kumar Das , Tomi Kinnunen This is my paper
classification eess.AScs.CLcs.SD
keywords speechdeepfakesourcetracingcross-lingualgeneralizationmultilingualbenchmarkLFCCECAPA-TDNNself-supervisedlearningtext-to-speechattributionMLAAD
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 introduces the first benchmark for multilingual speech deepfake source tracing: given a synthetic speech sample, identify which text-to-speech architecture produced it, in both monolingual and cross-lingual settings. It builds a linguistically balanced dataset from the MLAAD corpus covering six languages in three language families and four TTS architectures, with protocols for cross-lingual transfer, unseen languages, and unseen speakers. The central finding is a trade-off: language-fine-tuned self-supervised front-ends achieve the best monolingual accuracy (up to 97.91% macro-F1), while a DSP-based system—LFCC features with an ECAPA-TDNN backend—is the most robust cross-lingually (88.40% average macro-F1). The paper also shows that same-language-family transfer outperforms cross-family transfer, and that unseen-language generalization is strong for LFCC-based systems.

What carries the argument

The benchmark's engine is the MCL-MLAAD dataset: a balanced subset of the MLAAD corpus with six languages across three language families (Germanic, Romance, Slavic) and four TTS architectures (Griffin-Lim, Bark, XTTS v1.1, XTTS v2), each utterance available clean and with four noise perturbations from MUSAN. The central comparison is between DSP front-ends (LFCCs fed to ResNet18, AASIST, or ECAPA-TDNN) and SSL front-ends (XLS-R, wav2vec2.0, and language-fine-tuned variants fed to AASIST), evaluated under four protocols: mono/cross-lingual, language-family, leave-one-language-out, and seen/unseen pseudo-speakers.

What would settle it

Check the MCL-MLAAD dataset metadata to confirm that all four TTS architectures (Griffin-Lim, Bark, XTTS v1.1, XTTS v2) appear in each of the six languages. If any language lacks a class, recompute the cross-lingual macro-F1 scores using only the present classes; a large drop in LFCC-ECAPA-TDNN's 88.40% average would indicate the result partly reflects language detection rather than source tracing.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central discovery is that cross-lingual deepfake source tracing is achievable with current architectures, but the optimal model differs by scenario. In monolingual settings, fine-tuning an SSL front-end (XLS-R) on the target language gives the highest accuracy (97.91% macro-F1). In cross-lingual settings—training on one language and testing on another—the hand-crafted LFCC front-end paired with an ECAPA-TDNN backend outperforms all SSL variants, including multilingual pre-trained ones, by a large margin (88.40% vs. 59.75% for XLSR-AASIST). This suggests that DSP features capture language-agnostic generation artifacts, whereas SSL representations are biased by th

Load-bearing premise

The claim that cross-lingual performance reflects source tracing rather than language detection assumes that every one of the four TTS architectures is available in all six languages; if a language lacks one architecture, the 4-way classification task is not well-defined and results could be confounded.

Editorial extensions

If this is right

  • Forensic attribution of deepfake speech can work across languages, not just within a single training language.
  • A model trained on one language family transfers better within that family, so deployment should consider linguistic typology when data are scarce.
  • DSP-based systems are a stronger baseline for cross-lingual source tracing than SSL-based systems in the tested conditions, contrary to the usual deep-learning advantage.
  • Unseen-language generalization is achievable with LFCC+ResNet18 or LFCC+ECAPA-TDNN (around 96–97% macro-F1), enabling tracing for languages absent from training data.
  • The lack of a consistent seen/unseen speaker gap suggests the models are not obviously shortcutting on speaker identity, though the pseudo-speaker labels are approximate.

Reading between the lines

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

  • Editorial inference: the strong leave-one-language-out results imply that a small set of languages may suffice to train a universal source tracer for these four architectures; testing on typologically more distant languages (e.g., Mandarin or Arabic) would stress this claim.
  • Editorial inference: the cross-lingual advantage of LFCC over XLS-R suggests that generation artifacts are partly channel-like and language-independent; future SSL fine-tuning could target artifact-discrimination directly rather than language-matched features.
  • Editorial inference: if the four TTS architectures are not all present in every language, the cross-lingual results could conflate language detection with source tracing; checking the MCL-MLAAD metadata is a simple way to settle this.
  • Editorial inference: the pseudo-speaker clustering relies on embeddings from an off-the-shelf speaker model, so the unseen-speaker robustness finding is an upper bound on true speaker invariance; a dataset with ground-truth speaker identities would be needed to confirm it.
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

4 major / 6 minor

Summary. Motivated by the lack of multilingual source tracing for speech deepfakes, the authors construct MCL-MLAAD from MLAAD, covering six languages across three families (Germanic, Romance, Slavic) and four TTS architectures (Griffin-Lim, Bark, XTTS v1.1, XTTS v2). They define four evaluation protocols: monolingual, cross-lingual, cross-family, leave-one-language-out (unseen language), and pseudo-speaker seen/unseen. They compare three LFCC-based DSP models (AASIST, ResNet18, ECAPA-TDNN backends) with SSL front-end models (XLS-R, wav2vec2 Large, and six language-fine-tuned XLS-R variants, all with AASIST). Reported results: W2V2(xx)-AASIST achieves the best average monolingual macro-F1 (97.91%), LFCC-ECAPA-TDNN the best cross-lingual average (88.40%), same-family transfers outperform cross-family, LFCC models generalize better than XLSR to unseen languages and pseudo-speakers, and no consistent seen/unseen speaker gap is found. Code, dataset, and protocol are released.

Significance. The paper provides a useful and reproducible benchmark resource: public code/data/protocols, a clean four-class design, and a broad comparison across feature types, language families, and pretraining strategies. If the confounds are resolved, it would be a solid first step for multilingual source tracing. However, the headline DSP-versus-SSL conclusion is not identified by the current experimental matrix because front-end, backend, and Mixup are compounded; the per-language TTS composition needs explicit verification; and single-run macro-F1 numbers lack statistical support.

major comments (4)
  1. [Section 4.4 / Table 4 (RQ4)] The central claim that 'DSP architectures inherently prioritize language-agnostic patterns' is not supported. All SSL models use the AASIST backend and Mixup, while the LFCC family uses three different backends and no Mixup. In the backend-matched comparison, LFCC-AASIST is the worst cross-lingual model (54.46%), below XLSR-AASIST (59.75%), W2V2(xx)-AASIST (73.45%) and W2V2EN-AASIST (78.18%). The best DSP score (88.40%) comes from LFCC-ECAPA-TDNN, and no SSL-ECAPA-TDNN run is reported. The observed advantage may therefore be due to the ECAPA-TDNN backend or the absence/presence of Mixup rather than to LFCC as a DSP front-end. Please add an SSL-ECAPA-TDNN baseline, or rephrase the conclusion to describe the specific LFCC-ECAPA-TDNN configuration and remove the 'inherently' generalization.
  2. [Section 2.1 / Table 1] The benchmark defines a 4-class source-tracing task, and all cross-lingual and unseen-language comparisons rely on the class structure being the same in every language. The text states that six languages and four TTS architectures were selected, but it never explicitly says that each language contains all four architectures, and Table 1 only gives per-language utterance totals. If, for example, one architecture is absent from Russian (which has fewer samples, 1,200 vs 2,100), cross-lingual scores could reflect language/class prior differences rather than source-tracing ability. Please state the coverage explicitly and provide a language × TTS contingency table.
  3. [Sections 3.2.3 and 4, Tables 4-6] Every macro-F1 number is a single run. Tables 4-6 report 36 cells and averages with no confidence intervals, standard deviations, or significance tests. Some headline differences are small (e.g., 97.91% vs 97.78% monolingual average), while macro-F1 on roughly 1,000 test samples per language can vary by several points across seeds. Because several conclusions are rankings (best model, within-family advantage, SSL-versus-DSP), the absence of seed variability makes the comparisons unverifiable. Please report at least three seeds with mean±std, or apply significance tests.
  4. [Sections 2.2.4 and 4.7, Table 6] The seen/unseen speaker protocol is based on pseudo-speaker labels from unsupervised ECAPA2 clustering with an elbow-selected k=18 and per-language thresholds θ. The authors themselves note the lack of ground-truth speaker labels and the low speaker count. Given this, the conclusion that models show 'potential robustness to speaker variation' is fragile and should not be presented as a key finding. Please mark Table 6 as exploratory only and either validate the clustering with a small labeled subset or soften the conclusion accordingly. Also clarify the exact rule for classifying a pseudo-speaker as 'seen' vs 'unseen' (e.g., what happens when the prior is between 0 and θ).
minor comments (6)
  1. [Section 3.2.3] The macro-F1 formula, F1 = 2PR/(P+R) with macro-averaged P and R, is nonstandard; macro-F1 is typically the average of per-class F1 values. Please define and use the standard version or explain why this variant was chosen.
  2. [Section 4.4] 'When backend architectures are comparable (top four subtables)' is confusing; Table 4 has three SSL subtables and three DSP subtables, not four. Please correct the cross-reference.
  3. [Section 2.1] There are typos: 'Multil-lingual' and 'Lingual-Balanced' (likely 'linguistically balanced').
  4. [Section 4.7] 'unlike was hypothesized' should be 'contrary to the hypothesis'.
  5. [References] Some cited references (e.g., [7], [8], [25], [30], [31], [34]-[37], [67]) are not clearly connected to source tracing or the benchmark; trimming would improve readability.
  6. [Table 1] The meaning of '(×5)' should be explained in the caption; as written it is ambiguous whether the counts are utterances or audio files after perturbation.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the benchmark is an empirical measurement with independent external resources and no result reduces to its inputs.

full rationale

This paper is an empirical benchmark study, not a derivation. The central outputs are measured macro-F1 scores of various front-end/back-end combinations on a new protocol (MCL-MLAAD) constructed by selecting six languages and four TTS architectures from the external MLAAD corpus. There is no fitted parameter that is later relabeled as a prediction: the benchmark protocol, train/dev/test splits, and evaluation metrics are all defined independently of the reported findings. The pseudo-speaker labels in §2.2.4 are derived from an external ECAPA2 speaker encoder and spherical k-means clustering, and the paper explicitly acknowledges the limitation that 'the speaker labels were derived through clustering process' (§4.7) and that synthetic speech 'does not even have crisply-defined speaker identity' (§2.2.4). These are honest measurement choices, not circular reasoning. The SSL front-ends are externally pretrained models (wav2vec2, XLS-R) and the fine-tuned variants use external CommonVoice data, so the language-generalization results are not built from the conclusions they support. All self-citations are background literature and none carries a load-bearing argument; no uniqueness theorem or prior result by the authors is invoked to force the design. The apparent DSP-vs-SSL comparison is partially confounded by backend choice (all SSL models use AASIST, while the best DSP system uses ECAPA-TDNN), but the paper explicitly conditions its claim on the backend: 'DSP-based methods with robust backend ECAPA-TDNN exhibit superior cross-lingual stability' (§4.4), and the conclusion says 'LFCC features combined with ResNet or ECAPA-TDNN backends demonstrate superior cross-lingual generalization' (§5). That is an experimental design limitation, not a reduction of the result to its own assumptions. No derivation chain in the paper equates an output to an input by construction, so there is no circularity to report.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central benchmark depends on several hand-chosen protocol parameters and domain assumptions about TTS artifact transfer and speaker pseudo-labeling. None of these are derived from theory.

free parameters (5)
  • speaker seen/unseen threshold theta = 0.074 to 0.085 (Table 2)
    Chosen per held-out language to balance seen/unseen pseudo-speaker group sizes; affects the speaker generalization analysis.
  • number of speaker clusters k = 18
    Set via elbow criterion on spherical k-means over ECAPA2 embeddings; the choice changes seen/unseen speaker assignments.
  • TTS model selection = Griffin-Lim, Bark, XTTS v1.1, XTTS v2
    Four of 91 available architectures chosen by hand; determines the class space of the benchmark.
  • language selection = en, de, fr, it, pl, ru
    Two languages from each of three families; chosen for typological coverage but not all 38 MLAAD languages.
  • Mixup alpha = 0.5
    Hyperparameter for SSL model training; influences generalization but not the central benchmark claim.
assumptions (4)
  • domain assumption All four TTS architectures are available for all six languages
    Stated nowhere in Section 2.1. Needed for the 4-class task to be comparable across languages; if false, cross-lingual numbers are confounded with language detection.
  • domain assumption ECAPA2 embeddings cluster into meaningful pseudo-speaker identities
    Section 2.2.4 uses ECAPA2 embeddings and spherical k-means to create pseudo-speaker labels. If clusters capture TTS artifacts or language rather than speaker, the seen/unseen speaker analysis is invalid.
  • domain assumption MLAAD synthetic samples preserve source-model identity across languages
    The benchmark assumes each TTS model leaves transferable artifacts independent of language. Underlying the cross-lingual protocol.
  • standard math Macro-F1 based on macro-averaged precision and recall is a valid comparison metric
    Section 3.2.3, aligns with prior ST work [26], but does not report variance.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Multilingual Source Tracing of Speech Deepfakes: A First Benchmark." pith.science (2026). https://pith.science/paper/YM2TOJ4I

@misc{pith2026250804143,
  author       = {Pith},
  title        = {Pith review of: Multilingual Source Tracing of Speech Deepfakes: A First Benchmark},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YM2TOJ4I}},
  note         = {Machine review of arXiv:2508.04143}
}
read the original abstract

Recent progress in generative AI has made it increasingly easy to create natural-sounding deepfake speech from just a few seconds of audio. While these tools support helpful applications, they also raise serious concerns by making it possible to generate convincing fake speech in many languages. Current research has largely focused on detecting fake speech, but little attention has been given to tracing the source models used to generate it. This paper introduces the first benchmark for multilingual speech deepfake source tracing, covering both mono- and cross-lingual scenarios. We comparatively investigate DSP- and SSL-based modeling; examine how SSL representations fine-tuned on different languages impact cross-lingual generalization performance; and evaluate generalization to unseen languages and speakers. Our findings offer the first comprehensive insights into the challenges of identifying speech generation models when training and inference languages differ. The dataset, protocol and code are available at https://github.com/xuanxixi/Multilingual-Source-Tracing.

Figures

Figures reproduced from arXiv: 2508.04143 by the authors.

Figure 1
Figure 1. Illustration of mono-lingual and cross-lingual sce￾narios on source tracing systems. The English data-trained system works well with English data (left) but fails with other languages (right). created it (or, more practically, which generative architecture is a likely origin of the deepfake speech sample). In foren￾sic science, source tracing (ST) of evidentiary materials, in￾cluding telephone recordings [17], digit… view at source ↗
Figure 2
Figure 2. Overview of the multilingual source tracing (ST) benchmark framework. The framework evaluates model generalization across languages and speakers through four experimental scenarios: (1) Monolingual (same-language/family training and testing); (2) Crosslingual (train on one language/family, test on another language/family); (3) Unseen languages generalization (train on multiple languages, test on unseen language); (4… view at source ↗
Figure 3
Figure 3. Architecture Comparison of DSP-based and SSL￾based Source Tracing (ST) Models. Red components represent variable components that can be replaced or adjusted in differ￾ent scenarios, while white components represent fixed compo￾nents that remain consistent across the models. number of LFCCs and 399 is the number of frames. We im￾plemented two classifiers: ResNet185 and ECAPA-TDNN6 . 3.2.2. Models with SSL front-end E… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Heatmaps illustrating the Macro F1-score (%) com￾parison across XLSR-AASIST, LFCC-ResNet18, and LFCC￾ECAPA-TDNN models. Rows indicate source language fami￾lies used for training (G: Germanic, R: Romance, S: Slavic). Columns indicate target language families used for ev…

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Fake-Mamba: Real-Time Speech Deepfake Detection Using Bidirectional Mamba as Self-Attention's Alternative

    eess.AS 2025-08 unverdicted novelty 5.0 of 10

    Fake-Mamba reports EERs of 0.97%, 1.74%, and 5.85% on three speech deepfake benchmarks, but the provided full text is an unrelated paper, so the claims cannot be verified.

Reference graph

Works this paper leans on

73 extracted references · 64 canonical work pages · cited by 1 Pith paper

  1. [1]

    The National Security Agency (NSA), Federal Bureau of Investigation (FBI), and Department of Homeland Security (DHS) recently released a joint report 1

    Introduction Recent advances in Generative AI (GenAI) have resulted in an unprecedented surge in synthetic data generation. The National Security Agency (NSA), Federal Bureau of Investigation (FBI), and Department of Homeland Security (DHS) recently released a joint report 1. It warns that synthetic media, especially deep- fake content, is now spreading q...

    arXiv 2025

  2. [2]

    Multilingual Source Tracing Benchmark As shown in Figure 2, to comprehensively and fair eval- uate Multilingual ST model performance in both Mono- and Cross-Lingual scenarios, we propose the linguistically- balanced dataset MCL-MLAAD and establish novel protocols for thorough assessment of generalization capabilities. 2.1. Dataset Accurate evaluation of M...

    work page

  3. [3]

    ST Models This section introduces the models used for ST tasks

    Multilingual Source Tracing Methods 3.1. ST Models This section introduces the models used for ST tasks. We first study how different input features influence performance 4https://huggingface.co/Jenthe/ECAPA2 in both mono-lingual and cross-lingual settings. Based on the type of front-end feature, we group the models into two cate- gories: DSP-based and SS...

    work page

  4. [4]

    Results and Discussion 4.1. Mono-Lingual Performance (RQ1) As shown in the diagonal entries of Table 4, monolingual per- formance demonstrates that W2V2(xx)-AASIST achieves high- est macro-F1 score of 97.91%, indicating that language-specific fine-tuning enhance phonetic differentiation. Notably, LFCC- ECAPA-TDNN attains 97.78% (18.98% higher than LFCC- A...

    work page arXiv

  5. [5]

    Furthermore, we first explore the effects of unseen languages and speakers in ST tasks

    Conclusion In this work, we establish the first multilingual benchmark for speech deepfake source tracing, covering both monolingual and cross-lingual scenarios across six languages and two model cat- egories (DSP- and SSL-based models). Furthermore, we first explore the effects of unseen languages and speakers in ST tasks. Our findings reveal three key i...

    work page

  6. [6]

    V oxpopulitts: a large-scale multilingual tts corpus for zero-shot speech generation,

    W. Liu, J. Bai, X. Cheng, J. Zuo, Z. Jiang, S. Ji, M. Fang, X. Yang, Q. Yang, and Z. Zhao, “V oxpopulitts: a large-scale multilingual tts corpus for zero-shot speech generation,” in Proceedings of the 31st International Conference on Computational Linguistics , 2025, pp. 10 293–10 297

    work page 2025

  7. [7]

    Promptvc: Flexible stylistic voice conversion in latent space driven by natural language prompts,

    J. Yao, Y . Yang, Y . Leiet al., “Promptvc: Flexible stylistic voice conversion in latent space driven by natural language prompts,” in Proceedings of IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2024, pp. 10 571–10 575

    work page 2024

  8. [8]

    A survey on speech deep- fake detection,

    M. Li, Y . Ahmadiadli, and X. P. Zhang, “A survey on speech deep- fake detection,”ACM Computing Surveys, vol. 58, no. 3, pp. 1–42, 2025, just Accepted

    work page 2025

Show all 73 references
  1. [9]

    Scams and frauds in the digital age: Ml-based detection and prevention strategies,

    S. V . J. Kolupuri, A. Paul, R. S. Bhowmick, and et al., “Scams and frauds in the digital age: Ml-based detection and prevention strategies,” in Proceedings of the 26th International Conference on Distributed Computing and Networking, 2025, pp. 340–345

    2025

  2. [10]

    Ge-gnn: Gated edge-augmented graph neural network for fraud detection,

    W. Zhang and C. Luo, “Ge-gnn: Gated edge-augmented graph neural network for fraud detection,” IEEE Transactions on Big Data, vol. 11, no. 4, pp. 1664–1676, 2025

    2025

  3. [11]

    Crowd density estimation based on multi-level attention maps,

    B. Ding, R. Han, Z. Ma, and X. Xuan, “Crowd density estimation based on multi-level attention maps,” in 2021 IEEE 5th Informa- tion Technology,Networking,Electronic and Automation Control Conference (ITNEC), vol. 5, 2021, pp. 1759–1765

    2021

  4. [12]

    Frect: Frequency-augmented convolu- tional transformer for robust time series anomaly detection,

    W. Zhang, D. Xu, G. Yao, X. Lin, R. Guan, C. Du, R. Han, X. Xuan, and C. Luo, “Frect: Frequency-augmented convolu- tional transformer for robust time series anomaly detection,” in Advanced Intelligent Computing Technology and Applications , D.-S. Huang, W. Chen, Y . Pan, and H...

    2025

  5. [13]

    Decomposition-based multi-scale transformer framework for time series anomaly detection,

    W. Zhang and C. Luo, “Decomposition-based multi-scale transformer framework for time series anomaly detection,”Neural Networks, vol. 187, p. 107399, 2025. [Online]. Available: https:// www.sciencedirect.com/science/article/pii/S0893608025002783

    2025

  6. [14]

    Asvspoof 2021: To- wards spoofed and deepfake speech detection in the wild,

    X. Liu, X. Wang, M. Sahidullah et al. , “Asvspoof 2021: To- wards spoofed and deepfake speech detection in the wild,” IEEE/ACM Transactions on Audio, Speech, and Language Pro- cessing, vol. 31, pp. 2507–2522, 2023

    2021

  7. [15]

    Asvspoof 5: Crowdsourced speech data, deepfakes, and adversarial attacks at scale,

    X. Wang, H. Delgado, H. Tak et al., “Asvspoof 5: Crowdsourced speech data, deepfakes, and adversarial attacks at scale,” in Pro- ceedings of The Automatic Speaker Verification Spoofing Coun- termeasures Workshop (ASVspoof 2024). ISCA, 2024, pp. 1–8

    2024

  8. [16]

    Add 2022: The first audio deep synthesis detection challenge,

    J. Yi, R. Fu, J. Tao et al. , “Add 2022: The first audio deep synthesis detection challenge,” in Proceedings of IEEE Interna- tional Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2022, pp. 9216–9220

    2022

  9. [17]

    Add 2023: The second audio deep- fake detection challenge,

    J. Yi, J. Tao, R. Fu et al., “Add 2023: The second audio deep- fake detection challenge,” in Proceedings of CEUR Workshop , vol. 3597, 2023, pp. 125–130, iSSN: 1613-0073

    2023

  10. [18]

    Trident of poseidon: A generalized approach for detecting deepfake voices,

    T. P. Doan, H. Dinh-Xuan, T. Ryu et al., “Trident of poseidon: A generalized approach for detecting deepfake voices,” in Pro- ceedings of the 2024 on ACM SIGSAC Conference on Computer and Communications Security , ser. CCS ’24. ACM, 2024, pp. 2222–2235

    2024

  11. [19]

    Xlsr-mamba: A dual-column bidirec- tional state space model for spoofing attack detection,

    Y . Xiao and R. K. Das, “Xlsr-mamba: A dual-column bidirec- tional state space model for spoofing attack detection,” IEEE Sig- nal Processing Letters, vol. 32, pp. 1276–1280, 2025

    2025

  12. [20]

    Nes2net: A lightweight nested architecture for foundation model driven speech anti-spoofing,

    T. Liu, D.-T. Truong, R. K. Das, K. A. Lee, and H. Li, “Nes2net: A lightweight nested architecture for foundation model driven speech anti-spoofing,” 2025. [Online]. Available: https://arxiv.org/abs/2504.05657

    2025

  13. [21]

    Interpretable temporal class activation representation for audio spoofing detection,

    M. Li and X. P. Zhang, “Interpretable temporal class activation representation for audio spoofing detection,” inProceedings of In- terspeech, 2024, pp. 1120–1124

    2024

  14. [22]

    Speech watermarking: An approach for the forensic analysis of digital telephonic recordings,

    M. Faundez-Zanuy, J. J. Lucena-Molina, and M. Hagm ¨uller, “Speech watermarking: An approach for the forensic analysis of digital telephonic recordings,” Journal of Forensic Sciences , vol. 55, no. 4, pp. 1080–1087, 2018, impact Factor: 1.5 (2023)

    2018

  15. [23]

    Digital image foren- sics via intrinsic fingerprints,

    A. Swaminathan, M. Wu, and K. J. R. Liu, “Digital image foren- sics via intrinsic fingerprints,” IEEE Transactions on Information Forensics and Security, vol. 3, no. 1, pp. 101–117, March 2008

    2008

  16. [24]

    Forensic analysis and anonymization of printed documents,

    T. Richter, S. Escher, D. Sch ¨onfeld et al., “Forensic analysis and anonymization of printed documents,” in Proceedings of the 6th ACM Workshop on Information Hiding and Multimedia Security. ACM, 2018, pp. 127–138

    2018

  17. [25]

    Transformer for authenticating the source microphone in digital audio forensics,

    M. Qamhan, Y . A. Alotaibi, and S. A. Selouani, “Transformer for authenticating the source microphone in digital audio forensics,” Forensic Science International: Digital Investigation , vol. 45, p. 301539, 2023

    2023

  18. [26]

    Primek-net: Multi- scale spectral learning via group prime-kernel convolutional neu- ral networks for single channel speech enhancement,

    Z. Lin, J. Wang, R. Li, F. Shen, and X. Xuan, “Primek-net: Multi- scale spectral learning via group prime-kernel convolutional neu- ral networks for single channel speech enhancement,” in ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processi...

    2025

  19. [27]

    Guidelines on cell phone forensics,

    W. Jansen and R. Ayers, “Guidelines on cell phone forensics,” National Institute of Standards and Technology (NIST), NIST Special Publication 800-101, 2007, nIST SP 800-101 Rev. 1. [Online]. Available: https://csrc.nist.gov/publications/detail/sp/ 800-101/rev-1/final

    2007

  20. [28]

    Source cell-phone recognition from recorded speech using non-speech segments,

    C. Hanilc ¸i and T. Kinnunen, “Source cell-phone recognition from recorded speech using non-speech segments,” Digital Signal Processing, vol. 35, pp. 75–85, 2014. [Online]. Available: https:// www.sciencedirect.com/science/article/pii/S1051200414002565

    2014

  21. [29]

    Forensic analysis of phone call networks,

    S. Catanese, E. Ferrara, and G. Fiumara, “Forensic analysis of phone call networks,”Social Network Analysis and Mining, vol. 3, no. 1, pp. 15–33, March 2013

    2013

  22. [30]

    Addressing noise and stochasticity in fraud detection for service networks,

    W. Zhang, D. Xu, X. Xuan, L. Jiang, G. Yao, R. Han, X. Lang, and C. Luo, “Addressing noise and stochasticity in fraud detection for service networks,” 2025. [Online]. Available: https://arxiv.org/abs/2505.00946

    2025 arXiv

  23. [31]

    Source tracing of audio deepfake systems,

    N. Klein, T. Chen, H. Tak et al., “Source tracing of audio deepfake systems,” in Proceedings of Interspeech. ISCA, 2024, pp. 1100– 1104

    2024

  24. [32]

    Generalized source tracing: Detect- ing novel audio deepfake algorithm with real emphasis and fake dispersion strategy,

    Y . Xie, R. Fu, Z. Wen et al., “Generalized source tracing: Detect- ing novel audio deepfake algorithm with real emphasis and fake dispersion strategy,” inProceedings of Interspeech. ISCA, 2024, pp. 4833–4837

    2024

  25. [33]

    Neural codec source tracing: Toward comprehensive attribution in open-set condition,

    Y . Xie, X. Wang, Z. Wang et al., “Neural codec source tracing: Toward comprehensive attribution in open-set condition,” arXiv preprint arXiv:2501.06514, 2025, preprint

    2025 arXiv

  26. [34]

    Multilingual speech translation from efficient finetuning of pretrained models,

    X. Li, C. Wang, Y . Tang et al., “Multilingual speech translation from efficient finetuning of pretrained models,” in Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natu- ral Language Proce...

    2021

  27. [35]

    Solving the unsolvable: Translating case law in hong kong,

    K. kui Sin, X. Xuan, C. Kit, C. H. yan Chan, and H. H. kin Ip, “Solving the unsolvable: Translating case law in hong kong,”

  28. [36]

    Multi-scene ro- bust speaker verification system built on improved ecapa-tdnn,

    X. Xuan, R. Jin, T. Xuan, G. Du, and K. Xuan, “Multi-scene ro- bust speaker verification system built on improved ecapa-tdnn,” in 2022 IEEE 6th Advanced Information Technology, Electronic and Automation Control Conference (IAEAC ), 2022, pp. 1689–1693

    2022

  29. [37]

    Translaw: Benchmarking large language models in multi-agent simulation of the collaborative translation,

    X. Xuan, K. kui Sin, Y . Zhou, and C. Kit, “Translaw: Benchmarking large language models in multi-agent simulation of the collaborative translation,” 2025. [Online]. Available: https://arxiv.org/abs/2507.00875

    2025 arXiv

  30. [38]

    Probing the feasibility of mul- tilingual speaker anonymization,

    S. Meyer, F. Lux, and N. T. Vu, “Probing the feasibility of mul- tilingual speaker anonymization,” in Proceedings of Interspeech, 2024, pp. 4448–4452

    2024

  31. [39]

    Introducing multilingual pho- netic information to speaker embedding for speaker verification,

    Z. Song, L. He, P. Wang et al. , “Introducing multilingual pho- netic information to speaker embedding for speaker verification,” in Proceedings of IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2024, pp. 10 091–10 095

    2024

  32. [40]

    Research on front-end of asv system based on mel spectrum in noise scenario,

    X. Xuan, J. Dong, and T. Xuan, “Research on front-end of asv system based on mel spectrum in noise scenario,” in 2022 IEEE 10th Joint International Information Technology and Artificial In- telligence Conference (ITAIC), vol. 10, 2022, pp. 2638–2642

    2022

  33. [41]

    Research on acoustic feature extractor for automatic speaker verification systerm,

    X. Xuan and R. Han, “Research on acoustic feature extractor for automatic speaker verification systerm,” in 2022 IEEE 10th Joint International Information Technology and Artificial Intelligence Conference (ITAIC), vol. 10, 2022, pp. 2628–2633

    2022

  34. [42]

    Madd: A multi-lingual multi-speaker audio deepfake detection dataset,

    X. Qi, H. Gu, J. Yi et al., “Madd: A multi-lingual multi-speaker audio deepfake detection dataset,” in 2024 IEEE 14th Interna- tional Symposium on Chinese Spoken Language Processing (ISC- SLP). IEEE, 2024, pp. 466–470

    2024

  35. [43]

    Research on speaker identifica- tion models based on cnn and additive angular margin loss,

    X. Xuan, R. Han, and B. Ding, “Research on speaker identifica- tion models based on cnn and additive angular margin loss,” in 2021 2nd International Conference on Electronics, Communica- tions and Information Technology (CECIT), 2021, pp. 1046–1050

    2021

  36. [44]

    Conformer-based speaker recogni- tion model for real-time multi-scenarios,

    X. Xuan, R. Han, and J. Gao, “Conformer-based speaker recogni- tion model for real-time multi-scenarios,” Computer Engineering and Applications, vol. 60, no. 7, pp. 147–156, 2024

    2024

  37. [45]

    Scaling speech technology to 1,000+ languages,

    V . Pratap, A. Tjandra, B. Shi et al., “Scaling speech technology to 1,000+ languages,” Journal of Machine Learning Research , vol. 25, no. 97, pp. 1–52, 2024

    2024

  38. [46]

    Neural codec language models are zero-shot text to speech synthesizers,

    S. Chen, C. Wang, Y . Wu, Z. Zhang, L. Zhou, S. Liu, Z. Chen, Y . Liu, H. Wang, J. Liet al., “Neural codec language models are zero-shot text to speech synthesizers,” IEEE Transactions on Au- dio, Speech and Language Processing, 2025

    2025

  39. [47]

    Polyglotfake: A novel multilingual and multimodal deepfake dataset,

    Y . Hou, H. Fu, C. Chenet al., “Polyglotfake: A novel multilingual and multimodal deepfake dataset,” inInternational Conference on Pattern Recognition. Cham: Springer Nature Switzerland, 2024, pp. 180–193

    2024

  40. [48]

    Musan: A music, speech, and noise corpus,

    D. Snyder, G. Chen, and D. Povey, “Musan: A music, speech, and noise corpus,” arXiv preprint arXiv:1510.08484, 2015

    2015 arXiv

  41. [49]

    Mlaad: The multi- language audio anti-spoofing dataset,

    N. M. M ¨uller, P. Kawa, W. H. Choonget al., “Mlaad: The multi- language audio anti-spoofing dataset,” in2024 International Joint Conference on Neural Networks (IJCNN). IEEE, 2024, pp. 1–7

    2024

  42. [50]

    Heterogeneity over homogeneity: Investigating multilingual speech pre-trained models for detecting audio deepfake,

    O. C. Phukan, G. Kashyap, A. B. Buduru et al., “Heterogeneity over homogeneity: Investigating multilingual speech pre-trained models for detecting audio deepfake,” in Findings of the Asso- ciation for Computational Linguistics: NAACL 2024 , 2024, pp. 2496–2506

    2024

  43. [51]

    Transferring audio deepfake detection capability across lan- guages,

    Z. Ba, Q. Wen, P. Cheng, Y . Wang, F. Lin, L. Lu, and Z. Liu, “Transferring audio deepfake detection capability across lan- guages,” in Proceedings of the ACM Web Conference 2023, 2023, pp. 2033–2044

    2023

  44. [52]

    Towards quantifying and reducing language mismatch ef- fects in cross-lingual speech anti-spoofing,

    T. Liu, I. Kukanov, Z. Pan, Q. Wang, H. B. Sailor, and K. A. Lee, “Towards quantifying and reducing language mismatch ef- fects in cross-lingual speech anti-spoofing,” in2024 IEEE Spoken Language Technology Workshop (SLT). IEEE, 2024, pp. 1185– 1192

    2024

  45. [53]

    Are audio deepfake detection models polyglots?

    B. Marek, P. Kawa, and P. Syga, “Are audio deepfake detection models polyglots?” arXiv preprint arXiv:2412.17924, 2024

    2024 arXiv

  46. [54]

    Aasist: Audio anti-spoofing us- ing integrated spectro-temporal graph attention networks,

    J. Jung, H. S. Heo, H. Tak et al., “Aasist: Audio anti-spoofing us- ing integrated spectro-temporal graph attention networks,” inPro- ceedings of IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2022, pp. 6367–6371

    2022

  47. [55]

    Shortcut learning in deep neural networks,

    R. Geirhos et al. , “Shortcut learning in deep neural networks,” Nature Machine Intelligence, vol. 2, no. 11, pp. 665–673, 2020

    2020

  48. [56]

    Ecapa2: A hybrid neural net- work architecture and training strategy for robust speaker embed- dings,

    J. Thienpondt and K. Demuynck, “Ecapa2: A hybrid neural net- work architecture and training strategy for robust speaker embed- dings,” in 2023 IEEE Automatic Speech Recognition and Under- standing Workshop (ASRU). IEEE, 2023, pp. 1–8

    2023

  49. [57]

    Concept decompositions for large sparse text data using clustering,

    I. S. Dhillon and D. S. Modha, “Concept decompositions for large sparse text data using clustering,” Machine Learning , vol. 42, no. 1, pp. 143–175, 2001. [Online]. Available: https://doi.org/10.1023/A:1007612920971

    2001 doi

  50. [58]

    Review on determining number of cluster in k-means clustering,

    T. M. Kodinariya and P. R. Makwana, “Review on determining number of cluster in k-means clustering,” International Journal, vol. 1, no. 6, pp. 90–95, 2013

    2013

  51. [59]

    A comparison of features for synthetic speech detection,

    M. Sahidullah, T. Kinnunen, and C. Hanilc ¸i, “A comparison of features for synthetic speech detection,” in Proceedings of Inter- speech, 2015, pp. 2087–2091

    2015

  52. [60]

    Common V oice: A massively-multilingual speech corpus,

    R. Ardila, M. Branson, K. Davis et al. , “Common V oice: A massively-multilingual speech corpus,” in Proceedings of the Twelfth Language Resources and Evaluation Conference , 2020, pp. 4218–4222

    2020

  53. [61]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 770–778

    2016

  54. [62]

    ECAPA- TDNN: Emphasized channel attention, propagation and aggrega- tion in TDNN based speaker verification,

    B. Desplanques, J. Thienpondt, and K. Demuynck, “ECAPA- TDNN: Emphasized channel attention, propagation and aggrega- tion in TDNN based speaker verification,” in Proceedings of In- terspeech, 2020, pp. 3830–3834, iSSN 2308-457X

    2020

  55. [63]

    Unsu- pervised cross-lingual representation learning for speech recogni- tion,

    A. Conneau, A. Baevski, R. Collobert, and A. Mohamed, “Unsu- pervised cross-lingual representation learning for speech recogni- tion,” in Proceedings of Interspeech, 2021, pp. 2426–2430

    2021

  56. [64]

    V oxpopuli: A large-scale multilingual speech corpus for representation learning, semi- supervised learning and interpretation,

    C. Wang, M. Riviere, A. Lee et al., “V oxpopuli: A large-scale multilingual speech corpus for representation learning, semi- supervised learning and interpretation,” inProceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th Internat...

    2021

  57. [65]

    MLS: A large-scale multilin- gual dataset for speech research,

    V . Pratap, Q. Xu, A. Sriram et al., “MLS: A large-scale multilin- gual dataset for speech research,” in Proceedings of Interspeech, 2020, pp. 2757–2761

    2020

  58. [66]

    mixup: Beyond empirical risk minimization,

    H. Zhang, M. Cisse, Y . N. Dauphin, and D. Lopez-Paz, “mixup: Beyond empirical risk minimization,” in International Conference on Learning Representations (ICLR), 2018. [Online]. Available: https://arxiv.org/abs/1710.09412

    2018 arXiv

  59. [67]

    V oxlingua107: A dataset for spoken lan- guage recognition,

    J. Valk and T. Alum ¨ae, “V oxlingua107: A dataset for spoken lan- guage recognition,” inProceedings of the 2021 IEEE Spoken Lan- guage Technology Workshop (SLT). IEEE, 2021, pp. 652–658

    2021

  60. [68]

    Speech recog- nition and keyword spotting for low-resource languages: Babel project research at cued,

    M. J. F. Gales, K. M. Knill, A. Ragni et al. , “Speech recog- nition and keyword spotting for low-resource languages: Babel project research at cued,” in Proceedings of the Fourth Interna- tional Workshop on Spoken Language Technologies for Under- Resourced Languages (SLTU-201...

    2014

  61. [69]

    wav2vec 2.0: A framework for self-supervised learning of speech represen- tations,

    A. Baevski, Y . Zhou, A. Mohamed, and et al., “wav2vec 2.0: A framework for self-supervised learning of speech represen- tations,” Advances in Neural Information Processing Systems , vol. 33, pp. 12 449–12 460, 2020

    2020

  62. [70]

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

    V . Panayotov, G. Chen, D. Povey, and S. Khudanpur, “Lib- rispeech: An asr corpus based on public domain audio books,” in Proceedings of IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2015, pp. 5206– 5210

    2015

  63. [71]

    Towards generalisable and calibrated audio deepfake detection with self-supervised represen- tations,

    O. Pascu, A. Stan, D. Oneata et al., “Towards generalisable and calibrated audio deepfake detection with self-supervised represen- tations,” in Proceedings of Interspeech, 2024, pp. 4828–4832

    2024

  64. [73]

    Unlocking the full potential of separa- ble convolutions on tensor cores,

    A. Cui, C. Zhao, X. Deng, G. Jiang, Y . Yang, G. Yao, R. Han, W. Zhang, and X. Xuan, “Unlocking the full potential of separa- ble convolutions on tensor cores,” in International Conference on Intelligent Computing. Springer, 2025, pp. 39–50

    2025

  65. [2025]

    Available: https://arxiv.org/abs/2501.09444

    [Online]. Available: https://arxiv.org/abs/2501.09444

Pith tools

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