Pith. sign in

REVIEW 2 major objections 5 minor 48 references

A Comparative Study on Proactive and Passive Detection of Deepfake Speech

T0 review · 2 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read This paper proposes a unified evaluation framework that lets proactive watermarking models and passive deepfake detectors be compared head-to-head, and reports that watermarking models can reach 0% EER on clean ASVspoof 2019 and 2021 LA…

desk verdict A genuinely useful benchmark for comparing watermarking and deepfake detectors, with a clean score-mapping idea—but the proactive protocol embeds watermarks into fake speech, which flatters the 0% EER and muddies the robustness comparison. read the letter →

arxiv 2506.14398 v1 pith:F2NYS6ME submitted 2025-06-17 cs.SD

Chia-Hua Wu , Wanying Ge , Xin Wang , Junichi Yamagishi , Yu Tsao , Hsin-Min Wang This is my paper
classification cs.SD
keywords deepfakespeechdetectionaudiowatermarkingproactivedefensepassiveASVspoofequalerrorraterobustnessevaluationanti-spoofing
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's goal is to make proactive watermarking models and passive deepfake detectors directly comparable for binary real-vs-fake speech classification. It sets up a common protocol: all models train and test on the same ASVspoof LA datasets, all outputs get converted to a single realness score, and all are scored with equal error rate (EER). Under this protocol, the watermarking models Timbre and AudioSeal classify real and spoofed utterances perfectly on clean audio (0% EER), matching or beating the passive detectors AASIST and SSL-AASIST. Once transmissions and manipulations are applied, every model degrades; Timbre is on average the most robust, but codecs and pitch shift still break it. A careful reader would care because choosing a defense currently lacks a fair basis, and this framework is a first attempt at providing one.

What carries the argument

The load-bearing mechanism is the scoring formula in Eq. (1), which converts a watermarking model's per-bit logits into a single realness score without retraining: two bit-wise disjoint random messages $M_{real}$ and $M_{fake}$ are assigned to real and fake utterances, and the score is $s = \frac{1}{L} \sum_{l=1}^{L} (s_{w,l} q(m_{real,l}) - s_{w,l} q(m_{fake,l}))$ with $q(1)=1$, $q(0)=-1$, so that agreement with $M_{real}$ pushes the score up and agreement with $M_{fake}$ pushes it down. This, together with the shared evaluation protocol (same datasets, same EER metric, and a common set of transmissions and manipulations applied after watermarking), is what makes the head-to-head comparison possible.

What would settle it

Run the same clean-set evaluation but watermark only the real utterances with $M_{real}$ and leave all fake utterances unwatermarked, classifying an utterance as real only if the recovered message matches $M_{real}$; if the EER rises far above 0%, the paper's claim of perfect proactive detection under realistic conditions would be undermined.

Watch

Extended reading notes

Core claim

The central discovery, stated on the paper's own terms, is that under a unified evaluation protocol, proactive watermarking models can perfectly separate real from spoofed speech on the ASVspoof 2019 LA and ASVspoof 2021 LA evaluation sets, reaching 0% EER even for unseen speakers and unseen generators, and that no model, proactive or passive, remains reliable under all distorted channel and manipulation conditions tested. In particular, the paper finds Timbre the most robust on average (8.87% and 9.02% EER across distorted conditions), while AudioSeal collapses on similar codecs despite its training augmentation, and AASIST degrades sharply on the 2021 LA transmission-heavy set.

Load-bearing premise

The whole comparison rests on the assumption that a watermarking model can be evaluated as a deepfake detector by watermarking both real and fake utterances with different messages and classifying by which message is recovered; in actual proactive defense, only genuine content carries the defender's watermark, so this protocol may not reflect the real attack scenario.

Editorial extensions

If this is right

  • If the framework is correct, defense selection can be based on measured EER across a shared protocol rather than on community-specific benchmarks.
  • The perfect clean-set performance of watermarking models implies that when the defender controls both embedding and extraction, binary deepfake detection is solved in the clean case; the hard remaining problem is robustness.
  • Timbre's robustness to time-stretch but vulnerability to pitch shift and codecs indicates that frequency-domain watermarking has a specific weakness profile that should be addressed directly.
  • AudioSeal's failure on similar codecs despite augmentation suggests that model-specific augmentation does not generalize across codec families, so robustness must be evaluated rather than assumed.
  • All models degrade on the 2021 LA set, indicating that real-world transmission is the dominant challenge for deployment.
  • The protocol can be reused to benchmark future watermarking models and detectors on equal footing.
  • The reported EERs provide concrete baselines: Timbre at 8.87% and SSL-AASIST at 12.41% average on distorted 2019 LA data.

Reading between the lines

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

  • An implication the authors leave implicit is that the protocol's 0% EER depends on watermarking the fake utterances too; in a real proactive defense only genuine content carries a watermark, so an attacker who simply emits unwatermarked deepfakes would bypass the classifier, and the reported perfect separation may not transfer.
  • The scoring formula treats each bit's logit as evidence for one of two messages; a natural testable extension is to evaluate whether the choice of message pair (e.g., Hamming distance or bit length) changes the EER, since Eq. (1) should be invariant to message complement in theory but bit-level calibration may not be.
  • Because proactive and passive models fail on different distortions (e.g., Timbre on pitch shift, AASIST on time stretch), an implicit avenue is fusing both score types; the paper does not attempt this fusion.
  • The framework could be extended to a true scenario where fakes are not watermarked, likely changing the ranking of proactive versus passive methods and providing a more direct measure of real-world usability.
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

2 major / 5 minor

Summary. The paper proposes a unified evaluation protocol for comparing passive deepfake detectors (AASIST, SSL-AASIST) with proactive watermarking models (Timbre, AudioSeal) on the ASVspoof 2019 and 2021 LA datasets. In the protocol, real utterances are watermarked with message M_real and fake utterances with M_fake, and the bit-level watermark detector scores are merged via Eq. (1) into a continuous real-vs-fake score. The study reports clean-condition EERs and EERs under 17 transmission/manipulation conditions, concluding that watermarking models can achieve 0% EER in clean conditions and that Timbre is the most robust model on average under distortions.

Significance. If the proposed protocol corresponded to a realistic deployment scenario, the paper would provide a valuable common testbed for two research communities, and the open-source code and use of standard ASVspoof protocols are strengths. However, the central protocol is not a realistic proactive-defense threat model: fake utterances are watermarked by the defender, which an attacker would not do. The headline 0% EER and the robustness ranking are therefore not established for the claimed task, although the experimental infrastructure is potentially reusable after a threat-model correction.

major comments (2)
  1. [Section 3.2, Eq. (1)] The protocol watermarks both real and fake test utterances with defender-chosen messages and then classifies by which message is recovered. In any realistic proactive deployment, only genuine content is watermarked by the content owner; fake speech is generated by an attacker and arrives unwatermarked. The reported 0% EER in Section 4.2 and Table 1 therefore demonstrates watermark-message recovery on defender-watermarked fakes, not the ability to distinguish watermarked real speech from unwatermarked deepfake speech. Please add an evaluation condition in which only real utterances are watermarked and fake utterances are left unwatermarked, or explicitly scope all claims to the oracle-assisted protocol used here.
  2. [Section 4.2, Table 1] The conclusion that Timbre appears to be the most robust is based on the average EER over conditions that mix fully unseen and partially seen distortions. Because Timbre has no partially-seen asterisks in Table 1 while SSL-AASIST and AudioSeal do, the pooled average is not a like-for-like robustness measure. Please report averages separately for fully unseen conditions or provide a statistical comparison (for example, bootstrap confidence intervals over utterances) before ranking the models.
minor comments (5)
  1. [Abstract and Section 3.3] The abstract says the paper analyzes robustness against 'adversarial attacks', but all conditions in Section 3.3 are non-adaptive random distortions and transmissions rather than adversarially optimized perturbations. Please use 'distortions and transmissions' or another accurate term.
  2. [Section 3.2, Eq. (1)] The two messages are described as 'bit-wise disjoint', but the score formula in Eq. (1) gives zero contribution from bit positions where both M_real and M_fake contain 0. Using complementary messages (M_fake = bitwise NOT M_real) would avoid wasting watermark capacity; please clarify the message design or explain why this does not affect the results.
  3. [Section 4.2, Table 1] EER values above 50%, such as AudioSeal's 97.40% under DAC, mean that the score convention is effectively inverted for that condition; the same discrimination would give an EER below 50% if the decision rule were flipped. This does not change the qualitative fragility finding, but the interpretation should be stated more carefully.
  4. [Section 4.1] The passive detectors use released checkpoints while the proactive models are trained by the authors, but no training details such as number of epochs, learning rate, or hyperparameters are given in the paper. Since the code is available, please add a brief training setup description or point to a specific configuration file.
  5. [Section 4.1 and 3.4] For the ASVspoof 2021 LA 'clean' row in Table 1, the original corpus already contains lossy transmission channels in about 85% of the data. Please specify whether watermarking is applied before or after the channel degradation already present in the corpus, so the reader can interpret the 'None from Section 3.3' condition correctly.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the comparison uses independent pretrained models and external held-out ASVspoof data, and the proactive-model EERs are empirical watermark-extraction measurements rather than fitted parameters recycled as predictions.

full rationale

The paper's derivation chain is an evaluation protocol, not a parameter-fitting loop. Equation (1) is an explicitly defined score-fusion formula and does not encode the experimental outcome. All compared models are either official pretrained checkpoints (AASIST, SSL-AASIST) or models trained on ASVspoof 2019 LA train/dev (Timbre, AudioSeal) and then evaluated on held-out ASVspoof 2019/2021 test data. The reported 0% EER for proactive models is a measured consequence of the watermark detectors' bit-extraction accuracy on waveforms that the protocol watermarks with M_real or M_fake; the EER would be nonzero if extraction failed, and it is not obtained by fitting a parameter to the test labels. The choice to watermark fake utterances with M_fake is a threat-model simplification that limits external validity, but it is not a circular step: the paper does not assume the conclusion it reports, and no load-bearing claim is justified solely by a self-citation. Self-citations to ASVspoof and SSL-AASIST refer to external benchmarks and released checkpoints, not to unverified results that would be required for the comparison to go through.

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

The central claim rests on the choice of common datasets, the EER metric, and the assumption that watermarking fake speech with a known message yields a meaningful deepfake detection task. No extra free parameters or invented entities are introduced; all model weights come from prior training.

assumptions (3)
  • domain assumption EER is an appropriate common metric for comparing binary deepfake detection across passive detectors and watermarking models.
    Section 3.1. EER summarizes discriminative ability without a predefined threshold, but it assumes both model types produce a real-likelihood score. This is reasonable but still a modeling choice.
  • domain assumption A proactive watermarking model can be turned into a deepfake detector by watermarking real speech with M_real and fake speech with M_fake, then scoring with Eq. (1).
    Section 3.2. This is the load-bearing assumption; it presumes the defender can watermark the fake speech, which does not match typical proactive defense deployments where only genuine content is watermarked.
  • domain assumption The chosen transmission and manipulation conditions are sufficiently representative of real-world channel effects.
    Section 3.3. The list is arbitrary and partially reflects conditions seen in training of some models; the paper excludes some conditions but not others, which affects comparability.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Comparative Study on Proactive and Passive Detection of Deepfake Speech." pith.science (2026). https://pith.science/paper/F2NYS6ME

@misc{pith2026250614398,
  author       = {Pith},
  title        = {Pith review of: A Comparative Study on Proactive and Passive Detection of Deepfake Speech},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/F2NYS6ME}},
  note         = {Machine review of arXiv:2506.14398}
}
read the original abstract

Solutions for defending against deepfake speech fall into two categories: proactive watermarking models and passive conventional deepfake detectors. While both address common threats, their differences in training, optimization, and evaluation prevent a unified protocol for joint evaluation and selecting the best solutions for different cases. This work proposes a framework to evaluate both model types in deepfake speech detection. To ensure fair comparison and minimize discrepancies, all models were trained and tested on common datasets, with performance evaluated using a shared metric. We also analyze their robustness against various adversarial attacks, showing that different models exhibit distinct vulnerabilities to different speech attribute distortions. Our training and evaluation code is available at Github.

Figures

Figures reproduced from arXiv: 2506.14398 by the authors.

Figure 1
Figure 1. Overall comparison workflow between deepfake de￾tection and watermarking. The solid arrows (→) represent the deepfake detection process, while the dashed arrows (99K) in￾dicate the watermarking process. 2.2. Multi-bit watermarking Watermarking usually requires two components: a watermark embedder, which inserts a multi-bit message into the carrier waveform, and a watermark detector, which extracts and recon￾structs … view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

48 extracted references · 41 canonical work pages

  1. [1]

    Introduction Protection against artificially generated or manipulated deep- fake speech can be categorized into passive defense and proac- tive defense. Passive defense involves detecting whether speech from a content provider was generated by artificial intelligence, without any pre-processing of the input speech or prior assump- tions about the generato...

    work page 2019

  2. [2]

    A Comparative Study on Proactive and Passive Detection of Deepfake Speech

    Brief Review of Speech Deepfake Detection and Watermarking 2.1. Speech deepfake detection A speech deepfake detector acts as a binary classifier and de- cides whether the input waveform isreal or fake. 2 Given a waveformx∈ X, the detector functions asfd :X → Y, where Y≜{real,fake}is the set of output labels. In most cases, the detector is composed of a sc...

    work page Pith review arXiv 2025

  3. [3]

    1, upon which we compare deepfake detectors and watermarking mod- els for binary deepfake detection

    Evaluation of Speech Deepfake Detection and Watermarking We now explain the evaluation framework shown in Fig. 1, upon which we compare deepfake detectors and watermarking mod- els for binary deepfake detection. 3.1. Evaluation metrics For binary deepfake detection, we use EER as the main evalu- ation metric (§ 2.1). EER is a concise summary of a detector...

    work page

  4. [4]

    Experiment To assess the proposed evaluation framework, we conducted a comparative study using four representative DNN-based deep- fake detectors and watermarking models. 4.1. Experimental models Passive deepfake detectors: The two deepfake detectors we studied areAASIST[1] andSSL-AASIST[20]. AASIST is a state-of-the-art end-to-end (E2E) spoofing counterm...

    work page 2019

  5. [5]

    Conclusion In this work, we took the initiative to compare representative passive deepfake detectors (AASIST and SSL-AASIST) and proactive watermarking models (Timbre and AudioSeal) for the binary classification task of speech deepfake detection. Com- parisons based on unified evaluation metrics, scoring methods, diverse evaluation conditions, and common ...

    work page

  6. [6]

    This study was partially supported by JST AIP Acceleration Research (JP- MJCR24U3), MEXT KAKENHI Grant (24H00732), and JST PRESTO (JPMJPR23P9)

    Acknowledgements This work was conducted during the first author’s internship at the National Institute of Informatics (NII), Japan. This study was partially supported by JST AIP Acceleration Research (JP- MJCR24U3), MEXT KAKENHI Grant (24H00732), and JST PRESTO (JPMJPR23P9)

    work page

  7. [7]

    AASIST: Au- dio anti-spoofing using integrated spectro-temporal graph atten- tion networks,

    J.-w. Jung, H.-S. Heo, H. Tak, H.-j. Shimet al., “AASIST: Au- dio anti-spoofing using integrated spectro-temporal graph atten- tion networks,” inProc. ICASSP, 2022, pp. 6367–6371

    work page 2022

  8. [8]

    SLIM: Style- linguistics mismatch model for generalized audio deepfake detec- tion,

    Y . Zhu, S. Koppisetti, T. Tran, and G. Bharaj, “SLIM: Style- linguistics mismatch model for generalized audio deepfake detec- tion,” inProc. NeurIPS, 2024

    work page 2024

Show all 48 references
  1. [9]

    ASVspoof 2019: A large-scale public database of synthesized, converted and replayed speech,

    X. Wang, J. Yamagishi, M. Todisco, and Others, “ASVspoof 2019: A large-scale public database of synthesized, converted and replayed speech,”Computer Speech & Language, vol. 64, p. 101114, 2020

    2019

  2. [10]

    The codecfake dataset and countermeasures for the universally detection of deepfake audio,

    Y . Xie, Y . Lu, R. Fu, Z. Wenet al., “The codecfake dataset and countermeasures for the universally detection of deepfake audio,” IEEE Trans. Audio, Speech, Lang. Process., vol. 33, pp. 386–400, 2025

    2025

  3. [11]

    Detecting voice cloning attacks via Timbre watermarking,

    C. Liu, J. Zhang, T. Zhang, X. Yanget al., “Detecting voice cloning attacks via Timbre watermarking,” inProc. Netw. Distrib. Syst. Secur. Symp., 2024

    2024

  4. [12]

    Proac- tive detection of voice cloning with localized watermarking,

    R. S. Roman, P. Fernandez, H. Elsahar, A. D´efossezet al., “Proac- tive detection of voice cloning with localized watermarking,” in Proc. ICML, 2024

    2024

  5. [13]

    WavMark: Watermarking for audio generation,

    G. Chen, Y . Wu, S. Liu, T. Liuet al., “WavMark: Watermarking for audio generation,”arXiv preprint arXiv:2308.12770, 2023

    2023 arXiv

  6. [14]

    AudioMark- Bench: Benchmarking robustness of audio watermarking,

    H. Liu, M. Guo, Z. Jiang, L. Wanget al., “AudioMark- Bench: Benchmarking robustness of audio watermarking,” in Proc. NeurIPS Datasets and Benchmarks Track, 2024

    2024

  7. [15]

    One-class learning towards syn- thetic voice spoofing detection,

    Y . Zhang, F. Jiang, and Z. Duan, “One-class learning towards syn- thetic voice spoofing detection,”IEEE Signal Processing Letters, vol. 28, pp. 937–941, 2021

    2021

  8. [16]

    High fidelity neural audio compression,

    A. D ´efossez, J. Copet, G. Synnaeve, and Y . Adi, “High fidelity neural audio compression,”Transactions on Machine Learning Research, 2023

    2023

  9. [17]

    t-DCF: A detection cost function for the tandem assessment of spoof- ing countermeasures and automatic speaker verification,

    T. Kinnunen, K. A. Lee, H. Delgado, N. Evanset al., “t-DCF: A detection cost function for the tandem assessment of spoof- ing countermeasures and automatic speaker verification,” inProc. Odyssey, 2018, pp. 312–319

    2018

  10. [18]

    Defense against adversarial at- tacks on audio deepFake detection,

    P. Kawa, M. Plata, and P. Syga, “Defense against adversarial at- tacks on audio deepFake detection,” inProc. Interspeech, 2023, pp. 5276–5280

    2023

  11. [19]

    RawBoost: A raw data boosting and augmentation method applied to automatic speaker verification anti-spoofing,

    H. Tak, M. Kamble, J. Patino, M. Todiscoet al., “RawBoost: A raw data boosting and augmentation method applied to automatic speaker verification anti-spoofing,” inProc. ICASSP, 2022, pp. 6382–6386

    2022

  12. [20]

    ISO/IEC 30 107-1:2023, 2023

    International Organization for Standardization,Information Technology — Biometric Presentation Attack Detection — Part 1: Framework, International Organization for Standardization Std. ISO/IEC 30 107-1:2023, 2023. [Online]. Available: https: //www.iso.org/standard/83828.html

    2023

  13. [21]

    C. M. Bishop,Pattern Recognition and Machine Learning, ser. Information Science and Statistics. New York: Springer, 2006

    2006

  14. [22]

    Spoofing and countermeasures for speaker verification: A survey,

    Z. Wu, N. Evans, T. Kinnunen, J. Yamagishiet al., “Spoofing and countermeasures for speaker verification: A survey,”Speech Communication, vol. 66, pp. 130–153, 2015

    2015

  15. [23]

    Constant Q cepstral co- efficients: A spoofing countermeasure for automatic speaker ver- ification,

    M. Todisco, H. Delgado, and N. Evans, “Constant Q cepstral co- efficients: A spoofing countermeasure for automatic speaker ver- ification,”Computer Speech & Language, vol. 45, pp. 516–535, 2017

    2017

  16. [24]

    ASVspoof 2017: Automatic speaker verification spoofing and countermea- sures challenge evaluation plan,

    T. Kinnunen, N. Evans, J. Yamagishi, K. A. Leeet al., “ASVspoof 2017: Automatic speaker verification spoofing and countermea- sures challenge evaluation plan,” inProc. Interspeech, 2017, pp. 2–6

    2017

  17. [25]

    ASVspoof 2021: Towards spoofed and deepfake speech detection in the wild,

    X. Liu, X. Wang, M. Sahidullah, J. Patinoet al., “ASVspoof 2021: Towards spoofed and deepfake speech detection in the wild,”IEEE Trans. Audio, Speech, Lang. Process., vol. 31, pp. 2507–2522, 2023

    2021

  18. [26]

    Auto- matic speaker verification spoofing and deepfake detection using wav2vec 2.0 and data augmentation,

    H. Tak, M. Todisco, X. Wang, J.-w. Junget al., “Auto- matic speaker verification spoofing and deepfake detection using wav2vec 2.0 and data augmentation,” inProc. Odyssey, 2022, pp. 112–119

    2022

  19. [27]

    Spoofed training data for speech spoofing countermeasure can be efficiently created using neural vocoders,

    X. Wang and J. Yamagishi, “Spoofed training data for speech spoofing countermeasure can be efficiently created using neural vocoders,” inProc. ICASSP, 2023, pp. 1–5

    2023

  20. [28]

    Speech founda- tion model ensembles for the controlled singing voice deepfake detection (CtrSVDD) challenge 2024,

    A. Guragain, T. Liu, Z. Pan, H. B. Sailoret al., “Speech founda- tion model ensembles for the controlled singing voice deepfake detection (CtrSVDD) challenge 2024,” inProc. SLT, 2024, pp. 774–781

    2024

  21. [29]

    Speech watermarking with discrete intermediate representations,

    S. Ji, Z. Jiang, J. Zuo, M. Fanget al., “Speech watermarking with discrete intermediate representations,” inProc. AAAI, 2025

    2025

  22. [30]

    SilentCi- pher: Deep audio watermarking,

    M. K. Singh, N. Takahashi, W. Liao, and Y . Mitsufuji, “SilentCi- pher: Deep audio watermarking,” inProc. Interspeech, 2024, pp. 2235–2239

    2024

  23. [31]

    HiFi-GANw: Wa- termarked speech synthesis via fine-tuning of HiFi-GAN,

    X. Cheng, Y . Wang, C. Liu, D. Huet al., “HiFi-GANw: Wa- termarked speech synthesis via fine-tuning of HiFi-GAN,”IEEE Signal Processing Letters, vol. 31, pp. 2440–2444, 2024

    2024

  24. [32]

    Adversarial audio wa- termarking: Embedding watermark into deep feature,

    S. Wu, J. Liu, Y . Huang, H. Guanet al., “Adversarial audio wa- termarking: Embedding watermark into deep feature,” inProc. ICME, 2023, pp. 61–66

    2023

  25. [33]

    MaskMark: Robust neu- ralwatermarking for real and synthetic speech,

    P. O’Reilly, Z. Jin, J. Su, and B. Pardo, “MaskMark: Robust neu- ralwatermarking for real and synthetic speech,” inProc. ICASSP, 2024, pp. 4650–4654

    2024

  26. [34]

    D. A. Van Leeuwen and N. Br ¨ummer,An introduction to application-independent evaluation of speaker recognition sys- tems. Springer, 2007

    2007

  27. [35]

    Out of a hundred trials, how many errors does your speaker verifier make?

    N. Br ¨ummer, L. Ferrer, and A. Swart, “Out of a hundred trials, how many errors does your speaker verifier make?” inProc. In- terspeech, 2021, pp. 1059–1063

    2021

  28. [36]

    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

  29. [37]

    A study on data augmentation of reverberant speech for robust speech recog- nition,

    T. Ko, V . Peddinti, D. Povey, M. L. Seltzeret al., “A study on data augmentation of reverberant speech for robust speech recog- nition,” inProc. ICASSP, 2017, pp. 5220–5224

    2017

  30. [38]

    Pedalboard,

    P. Sobot, “Pedalboard,” Jul. 2021. [Online]. Available: https: //doi.org/10.5281/zenodo.7817838

    2021 doi

  31. [39]

    Definition of the opus audio codec,

    J.-M. Valin, K. V os, and T. Terriberry, “Definition of the opus audio codec,” Tech. Rep., 2012

    2012

  32. [40]

    High- fidelity audio compression with improved RVQGAN,

    R. Kumar, P. Seetharaman, A. Luebs, I. Kumaret al., “High- fidelity audio compression with improved RVQGAN,” inProc. NeurIPS, vol. 36, 2023, pp. 27 980–27 993

    2023

  33. [41]

    WavTokenizer: An effi- cient acoustic discrete codec tokenizer for audio language model- ing,

    S. Ji, Z. Jiang, W. Wang, Y . Chenet al., “WavTokenizer: An effi- cient acoustic discrete codec tokenizer for audio language model- ing,” inProc. ICLR, 2025

    2025

  34. [42]

    PyTorch: An im- perative style, high-performance deep learning library,

    A. Paszke, S. Gross, F. Massa, A. Lereret al., “PyTorch: An im- perative style, high-performance deep learning library,” inProc. NeurIPS, 2019

    2019

  35. [43]

    Finding, visualiz- ing, and quantifying latent structure across diverse animal vo- cal repertoires,

    T. Sainburg, M. Thielk, and T. Q. Gentner, “Finding, visualiz- ing, and quantifying latent structure across diverse animal vo- cal repertoires,”PLoS Computational Biology, vol. 16, no. 10, p. e1008228, 2020

    2020

  36. [44]

    timsainb/noisereduce: v1.0,

    T. Sainburg, “timsainb/noisereduce: v1.0,” Jun. 2019. [Online]. Available: https://doi.org/10.5281/zenodo.3243139

    2019 doi

  37. [45]

    DeepFilterNet: A low complexity speech enhancement frame- work for full-band audio based on deep filtering,

    H. Schr ¨oter, A. N. Escalante-B., T. Rosenkranz, and A. Maier, “DeepFilterNet: A low complexity speech enhancement frame- work for full-band audio based on deep filtering,” inProc. ICASSP, 2022, pp. 7407–7411

    2022

  38. [46]

    DeepFilterNet: Perceptually motivated real-time speech en- hancement,

    H. Schr ¨oter, T. Rosenkranz, A. N. Escalante-B., and A. Maier, “DeepFilterNet: Perceptually motivated real-time speech en- hancement,” inProc. Interspeech, 2023, pp. 2008–2009

    2023

  39. [47]

    librosa: Audio and music signal analysis in python,

    B. McFee, C. Raffel, D. Liang, D. P. Elliset al., “librosa: Audio and music signal analysis in python,” inProc. SciPy, 2015, pp. 18–25

    2015

  40. [48]

    wav2vec 2.0: A framework for self-supervised learning of speech representa- tions,

    A. Baevski, Y . Zhou, A. Mohamed, and M. Auli, “wav2vec 2.0: A framework for self-supervised learning of speech representa- tions,” inProc. NeurIPS, 2020

    2020

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.