A Fourier Explanation of AI-music Artifacts

Darius Afchar; Gabriel Meseguer-Brocal; Kamil Akesbi; Romain Hennequin

REVIEW 4 major objections 5 minor 45 references

A Fourier Explanation of AI-music Artifacts

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A Fourier proof shows AI music is detectable because deconvolution layers stamp predictable spectral peaks into every generated clip.

desk verdict Solid mechanistic account of transposed-convolution spectral artifacts, but the abstract's 'mathematically prove' overstates what is actually proven. read the letter →

arxiv 2506.19108 v1 pith:XPO222W5 submitted 2025-06-23 cs.SD

Darius Afchar , Gabriel Meseguer-Brocal , Kamil Akesbi , Romain Hennequin This is my paper · ORCID

classification cs.SD

keywords AImusicdetectiondeconvolutionartifactsFourieranalysisspectralpeakscheckerboardartifactarchitecture-dependentinterpretablegenerativeaudio

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

This paper tries to establish that the upsampling layers inside neural audio generators leave a permanent, detectable signature in the frequency spectrum of the music they produce: a row of small peaks whose positions are fixed by the architecture's stride sizes and do not depend on training data or learned weights. The argument is a Fourier-domain proof: zero-insertion upsampling periodizes a layer's spectrum, and if hidden layers carry a strong DC or low-frequency component from biases and ReLU activations, that component is cloned at predictable frequencies. The authors then show the pattern recurs through stacked deconvolution layers, producing a fractal-like set of peaks, and that a simple linear classifier reading this artifact fingerprint can match deep-learning detectors, exceeding 99% accuracy on several tests. If true, this gives an interpretable, architecture-level explanation for why synthetic music is detectable and why detectors transfer across models sharing the same deconvolution structure.

What carries the argument

The central mechanism is the Fourier transform's periodization identity: multiplying a signal by a Dirac comb samples it in time and copies its spectrum at intervals in frequency. The paper reinterprets zero-insertion upsampling as such a comb multiplication, so each deconvolution layer tiles copies of its input spectrum. The artifact fingerprint, defined as the average spectrum after subtracting local minima over a sliding window and restricted to a high-frequency band, is the operational object used to read the resulting peaks. It is the quantity the linear detector acts on.

What would settle it

Construct or train an audio generator whose deconvolution kernels are constant over the stride length, so their frequency response cancels the replicated peaks, while hidden layers are kept free of a DC bias; if the predicted peaks at $n f_s$ still appear in the average spectrum, the theory as stated is incomplete, and if they vanish, the artifact is not strictly unavoidable.

Watch

Extended reading notes

Core claim

Stated on the paper's own terms: a strided deconvolution is a zero-upsampling followed by a 1-strided convolution; in Fourier terms the zero-upsampling periodizes the spectrum, replicating the input spectrum at copies separated by the original sampling frequency $f_s$. Because biases and ReLU activations concentrate energy at $0$ Hz, the periodized copies appear as peaks at frequencies $n f_s$ for $n \leq \lfloor k/2 \rfloor$, up to the new Nyquist frequency $k f_s/2$. Through several deconvolution layers, these peaks clone recursively, so the final output contains $\lfloor P_{\max}/2\rfloor+1$ peaks where $P_{\max}=\prod_i k^{(i)}$ is the product of the strides; Encodec's strides $\{8,5,4,2\}$ yield 161 predicted peaks. The paper validates this with experiments on DAC, Encodec, and Musika!, and on Suno and Udio, and shows that the peak locations are the same for the same architecture trained on different datasets with different random seeds.

Load-bearing premise

The load-bearing assumption is that real generators' hidden layers have a strong low-frequency or DC component that survives both the learned convolution kernels and the nonlinearities; if a generator learned kernels that cancel the periodized peaks, the artifact and the detector would disappear.

Editorial extensions

If this is right

Detection is possible from a few engineered features: a logistic regression over the artifact fingerprint gives roughly 99% accuracy on open-source codecs and on commercial generators such as Suno v3.5 and Udio 130.
A detector trained on one model or dataset transfers to another model that shares the same deconvolution strides, because the peak positions are determined by the architecture alone.
The artifact pattern encodes the stride hyperparameters of the generator, so the recursive peak count can be read off as an architecture fingerprint.
The artifact cannot be removed by better training data or longer training; it is inherent to the use of deconvolution layers, and only an architectural change or carefully chosen kernels could eliminate it.
The Fourier analysis applies to the deconvolution layer itself, so the phenomenon is not specific to music: any generative model built on deconvolutions is predicted to carry a similar spectral fingerprint in its output.

Reading between the lines

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

If this account is right, a model that avoids detectable peaks would have to replace deconvolutions with other upsampling methods or learn kernels whose frequency response vanishes at the replicated peak positions—a concrete architectural target for future generators.
An adversary could in principle spoof the fingerprint by injecting matching peaks into real audio, which suggests that peak-presence alone may weaken as a detector over time and that phase or higher-order statistics may become necessary.
The failure on the unseen Udio 32 version suggests zero-shot detector transfer is really an architecture-matching problem; a testable prediction is that transfer will succeed exactly when two generators share the same deconvolution strides and fail when they do not.

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.

Desk Editor's Note

read the letter

Short version: this paper gives the clearest account yet of why transposed-convolution generators leave spectral peaks, and a 10K-parameter logistic regression that exploits them is genuinely competitive on open-source codecs. But the abstract's 'mathematically prove' is stronger than what Section 3 establishes: zero-insertion periodization is proven, survival through the convolution kernel and nonlinearities is argued heuristically and then demonstrated empirically. That distinction matters, though it doesn't sink the paper.

What's new: the recursive multi-layer peak-count formula (product of strides) that predicts, say, 161 peaks for Encodec's {8,5,4,2} strides; the demonstration that peak positions are stable across training seeds and datasets; and the framing of detector failure modes (resampling, pitch shift) that follows from the theory. The Fourier derivation itself is textbook — spectral imaging from zero-upsampling is classical — and they cite the prior checkerboard/audio artifact literature. The credit goes to making the prediction concrete and testing it on real models, with a public code repo.

Where it's soft: the persistence assumption. Section 3.2 proves zero-upsampling periodizes the spectrum, but the 1-strided convolution multiplies by the kernel's frequency response; a constant kernel of the right size would null all non-DC replicas. The paper acknowledges this and says small kernels are 'unlikely' to remove all artifacts — that's a heuristic, not a proof. Section 3.4 extends the caveat to nonlinearities, skip connections, batch norm. The empirical evidence on Encodec, DAC, Musika! shows the peaks actually survive in those trained checkpoints, but it doesn't close the logical gap. The closed-source comparison is also explicitly unfair (real tracks not released), so the 'on par' with SpecTTTra there is weaker than the Table 1 results. The Udio 32 zero-shot failure is a real bound.

Bottom line: this is a useful, mechanically grounded paper that deserves a serious referee. I'd recommend accepting it for review with a request to soften the proof language in the abstract and Section 3.2, and to move the persistence claim from 'proven' to 'predicted and empirically confirmed on the studied architectures.' The core contribution — architecture-dependent, data/weight-independent peak positions — holds up for the open models, and the detector is a nice byproduct. I'd bring it to reading group and cite it if I work in that area.

Referee Report

4 major / 5 minor

Summary. This paper studies spectral artifacts in AI-music generation. It shows that the zero-upsampling step of a deconvolution (transposed convolution) with stride k periodizes the hidden-layer spectrum, so that a DC or low-frequency component is replicated at multiples of the input sampling rate, producing predictable spectral peaks. The authors extend this to sequential CNNs with a recursive count of peaks and use the predicted peak positions to build a linear logistic-regression detector on an 'artifact fingerprint.' Experiments on Encodec, DAC, Musika!, Suno, and Udio are reported, with detection accuracy above 99% in several settings. The manuscript claims that the artifacts are architecture-dependent and independent of training data and learned weights.

Significance. The core observation is interesting and timely: if valid, it provides an interpretable, parameter-free explanation of why AI-music detectors work, and it predicts that artifact locations are fixed by stride hyperparameters rather than by training data. The paper is commendably concrete: the code repository is referenced, the peak positions are derived rather than fit, and the detector is evaluated on held-out splits. The strongest contribution is the framework connecting transposed-convolution arithmetic to Fourier periodization, together with the demonstration that a simple linear model can rival black-box detectors in some settings. However, the theoretical claim currently outruns the proof: only the zero-upsampling periodization is proven, while survival through kernels and nonlinearities is treated heuristically, and the closed-source parity claim is based on an unmatched evaluation protocol. These issues are load-bearing for the abstract's central claim, though they are addressable within the manuscript's scope.

major comments (4)

[Abstract; §3.2; §3.4] The abstract states that the paper 'mathematically prove[s] that their outputs exhibit systematic frequency artifacts,' but the formal derivation in §3.2 establishes periodization only for the zero-upsampling stage of a deconvolution. The subsequent 1-strided convolution multiplies the periodized spectrum by the kernel's frequency response, and the manuscript itself concedes that a constant kernel of size equal to the stride would cancel all of the predicted peaks except DC, answering only that small kernels are 'unlikely to remove all those artifacts entirely.' Section 3.4 explicitly states that persistence through nonlinear activations, skip connections, and batch normalization is not covered by the theory. Since the central claim and the detector's rationale require the peaks to be present in the final output, the authors should either prove a persistence result under precise conditions or explicitly downgrade the claim to a theorem about the intermediate representation plus an empirical persistence hypothesis. This is the main load-bearing gap, and it affects the title-level claim.
[§4.2, Fig. 4] The conclusion that the artifacts are 'solely architecture dependent' goes beyond the evidence presented. Figure 4 shows the same peak positions across four DAC models trained with different seeds and datasets, which is good evidence of position invariance for one architecture family, but it does not establish that artifacts are independent of learned weights in general; the peak amplitudes differ between runs, and only one model configuration is tested. To support the stronger claim, the authors should either verify position invariance on at least one additional architecture under multiple seeds, or phrase the claim as invariance of peak locations rather than of the artifacts themselves.
[§4.3, Table 2] The claimed parity with SpecTTTra-α in Table 2 is not supported by a matched evaluation. As the authors note, the SONICS real-audio tracks were not available and FMA tracks resampled to 16 kHz were used instead, so the scores are not directly comparable with the numbers reported in [16]. The abstract's statement that the method is 'on par with deep learning-based approaches' and the conclusion's 'on par with previous million-parameter models' therefore overstate the evidence for the closed-source scenario. The authors should either run the detector under the original SONICS protocol or restrict the parity claim to the open-source comparison in Table 1, where the baseline was evaluated under the same protocol.
[§4.3, Table 1] The open-source detection experiment uses auto-encoded real FMA audio as the 'synthetic' class rather than outputs of an end-to-end generative model. This is a reasonable codec-artifact study, and it matches the protocol of [17], but the abstract and introduction frame the result as AI-music detection. The gap between codec reconstruction detection and generated-music detection should be acknowledged explicitly in the claims, even though the Suno and Udio experiments partially bridge it, because the transfer from codec artifacts to full generators is itself an assumption rather than a proven fact.

minor comments (5)

[§3.2] The sentence 's and v have exactly the sample spectrum' appears to mean 'same spectrum'; please fix the typo and clarify that the equality holds for the continuous-time Fourier transforms, while the discrete-time spectra differ because of the change in sampling rate.
[§3.3] The notation in the recursive count is hard to follow: layer indices appear both as subscripts and superscripts, and the claim that a spectrum is 'cloned k(i+1)/2 times ... creating k(i+1) half-spectra' should be expanded with a two-layer worked example so that the product formula P_max is transparent.
[Fig. 3 caption] The phrase 'successive 213 samples' should be typeset as 2^13 samples or explained in words; as printed it is ambiguous.
[§4.2] The artifact fingerprint is defined as the average spectrum minus a sliding-window local minimum; the window size and bandwidth choices should be given numerical values, since the detector's behavior and the interpretation as a 'fingerprint' depend on them.
[Table 2] The dagger symbols for versions unseen during training are defined only in the text after the table; please move the definition into the table caption or a footnote so that the table is self-contained.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: peak locations are derived from architecture strides via Fourier identities, and detector accuracy is evaluated on held-out splits.

full rationale

The paper's derivation chain is self-contained rather than circular. Section 3.2 derives the artifact positions from the zero-insertion view of transposed convolutions, using standard Fourier identities (F[XT] is a comb, convolution becomes multiplication). The predicted peaks at n·fs depend only on the stride k and the sampling rate; they are not fitted parameters. The empirical checks (Figures 3 and 4) measure average spectra of independently trained checkpoints at the theoretically predicted frequencies, and the detector is a linear/logistic model evaluated on held-out splits against prior published baselines, so test accuracy is not forced by construction. The only genuinely open point is whether the periodized peaks survive the layer's convolution kernel and subsequent nonlinearities; the paper explicitly flags this as a heuristic or empirical question (Section 3.2: 'unlikely to remove all those artifacts entirely'; Section 3.4: 'we have not discussed whether non-linear activations allow these artifacts to persist'), and validates survival empirically. This is an acknowledged proof gap, not a circular derivation. Self-citations ([17], and the Deezer press release [1]) are used for context, dataset/methodology alignment, and comparison baselines; they are not the load-bearing justification for the artifact theorem.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The theory draws on standard Fourier and DSP results and the standard algebraic decomposition of transposed convolutions. It then adds two domain assumptions (DC-rich hidden features; short kernels cannot erase replicas) and a sequential-CNN induction that is only sketched. The detector adds fitted logistic weights and hand-set fingerprint windows. No new physical entities are postulated.

free parameters (2)

Linear detector weights (logistic regression) = learned on training splits, not reported explicitly
The claimed 99% detection accuracy comes from a supervised logistic regression on the artifact fingerprint; the weights are fitted to data, not derived from the theory.
Artifact fingerprint hyperparameters = sliding-window size and bandwidths [5kHz, 16kHz] or [1kHz, 8kHz]
Chosen by hand to emphasize local peaks and discard melodic content and codec cutoff; these choices are data-motivated and are not part of the theoretical derivation.

assumptions (5)

standard math Fourier duality: convolution in time is multiplication in frequency; Dirac-comb sampling leads to periodic summation of the spectrum.
Used throughout Section 3 to derive the periodization effect; the paper cites [31] for this background.
domain assumption A strided transposed convolution equals zero-upsampling followed by a 1-strided convolution.
Applied in Sections 3.1 and 3.2; standard result from [32], and the basis for modeling deconvolution as zero-insertion.
domain assumption Hidden layers carry a strong DC or low-frequency component, from additive biases and ReLU activations, so the periodized copies appear as peaks.
Invoked in Section 3.2 to claim peaks at integer multiples of fs; plausible for common networks but not proven for all architectures.
ad hoc to paper Learned kernels are unlikely to cancel the replicated peaks because kernel sizes are much smaller than the signal.
Section 3.2 states that 'limited expressivity' makes it unlikely that all artifacts are removed; this heuristic bridges the proven periodization step to the claimed artifacts in outputs.
ad hoc to paper Multi-layer artifacts follow a sequential induction that replicates previous peaks in a countable pattern, giving P_max as the product of strides.
Section 3.3 presents this for simple sequential CNNs; effects of nonlinearities, skip connections, and batch normalization are deferred to experiments.

how reviews work

0 comments

Cite this review

Pith. "Pith review of A Fourier Explanation of AI-music Artifacts." pith.science (2026). https://pith.science/paper/XPO222W5

@misc{pith2026250619108,
  author       = {Pith},
  title        = {Pith review of: A Fourier Explanation of AI-music Artifacts},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XPO222W5}},
  note         = {Machine review of arXiv:2506.19108}
}

read the original abstract

The rapid rise of generative AI has transformed music creation, with millions of users engaging in AI-generated music. Despite its popularity, concerns regarding copyright infringement, job displacement, and ethical implications have led to growing scrutiny and legal challenges. In parallel, AI-detection services have emerged, yet these systems remain largely opaque and privately controlled, mirroring the very issues they aim to address. This paper explores the fundamental properties of synthetic content and how it can be detected. Specifically, we analyze deconvolution modules commonly used in generative models and mathematically prove that their outputs exhibit systematic frequency artifacts -- manifesting as small yet distinctive spectral peaks. This phenomenon, related to the well-known checkerboard artifact, is shown to be inherent to a chosen model architecture rather than a consequence of training data or model weights. We validate our theoretical findings through extensive experiments on open-source models, as well as commercial AI-music generators such as Suno and Udio. We use these insights to propose a simple and interpretable detection criterion for AI-generated music. Despite its simplicity, our method achieves detection accuracy on par with deep learning-based approaches, surpassing 99% accuracy on several scenarios.

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

45 extracted references · 36 canonical work pages

[16]

Provocations from the Humanities for Generative AI Research,

L. Klein, M. Martin, A. Brock, M. Antoniak, M. Walsh, J. M. Johnson, L. Tilton, and D. Mimno, “Provocations from the Humanities for Generative AI Research,” arXiv preprint arXiv:2502.19190, 2025

arXiv 2025
[17]

AI Art is Theft: Labour, Extraction, and Exploitation: Or, On the Dangers of Stochastic Pol- locks,

T. S. Goetze, “AI Art is Theft: Labour, Extraction, and Exploitation: Or, On the Dangers of Stochastic Pol- locks,” in ACM FAccT, 2024

work page 2024
[1]

It’s not really enjoyable to make music now [...] I think the majority of people don’t enjoy the majority of the time they spend making music

INTRODUCTION “It’s not really enjoyable to make music now [...] I think the majority of people don’t enjoy the majority of the time they spend making music. ”— M. Shulman, CEO at Suno. Meanwhile, millions of users seem to enjoy creating AI-generated music. As a result, it was recently reported that at least a fifth of music delivered to streaming plat- fo...

work page 2025
[2]

pitched noise

RELATED WORK 2.1 AI-music detection Mirroring the recent boom of commercial AI-music gener- ation services (e. g.,Suno, Udio [16]), the task of AI-music detection is relatively novel. Only a few works have been published so far [14, 16, 17]. These early works propose CNN-like models to learn to classify real and synthetic sig- nals. They discuss several c...

arXiv 2025
[3]

de- convolution

FOURIER ANALYSIS OF ARTIFACTS In this work, we propose to reinterpret Convolutional Neu- ral Networks ( CNN) under the lenses of Fourier trans- forms. Instead of viewing inputs and outputs and hidden layers as time-based signals, we analyze their frequency- based decompositions (i. e., spectra). In the field of MIR, it is frequent to use spectrograms to p...

work page
[4]

EXPERIMENT In this section, we validate our theoretical findings and close the gaps left out by theory with empirical results. Our research questions are the following: RQ1 Are the artifacts solely architecture dependent? RQ2 Can they be used to detect synthetic music? We show that the models we study (both open-source and closed-source) exhibit peaks in ...

work page
[5]

CONCLUSION We propose a theoretical analysis for the emergence of peak artifacts in AI-music, formalize their recursive shape, and predict their architecture independence. Our exper- iments confirmed this latter observation, and we have used our new-found knowledge to craft a simple detector with performances on par with previous million-parameter models....

work page
[6]

Deezer deploys cutting-edge AI detec- tion tool for music streaming,

Deezer, “Deezer deploys cutting-edge AI detec- tion tool for music streaming,” https://newsroom- deezer.com/2025/04/deezer-reveals-18-of-all-new- music-uploaded-to-streaming-is-fully-ai-generated/, 2025, [Online; accessed 22-March-2025]

work page 2025

Show all 45 references

[7]

Watching the gen- erative AI hype bubble deflate,

D. G. Widder and M. Hicks, “Watching the gen- erative AI hype bubble deflate,” arXiv preprint arXiv:2408.08778, 2024

2024 arXiv
[8]

Be- hind the Deepfake: 8% Create; 90% Concerned. Sur- veying public exposure to and perceptions of deepfakes in the UK,

T. Sippy, F. Enock, J. Bright, and H. Z. Margetts, “Be- hind the Deepfake: 8% Create; 90% Concerned. Sur- veying public exposure to and perceptions of deepfakes in the UK,” arXiv preprint arXiv:2407.05529, 2024

2024 arXiv
[9]

CEO of AI Music Company Says People Don’t Like Making Music,

404 Media, “CEO of AI Music Company Says People Don’t Like Making Music,” https://www.404media.co/ceo-of-ai-music-company- says-people-dont-like-making-music/, 2025, [Online; accessed 13-June-2025]

2025
[10]

Music labels sue AI song gener- ators Suno and Udio for copyright infringement,

The Guardian, “Music labels sue AI song gener- ators Suno and Udio for copyright infringement,” https://www.theguardian.com/music/article/2024/ jun/25/record-labels-sue-ai-song-generator-apps- copyright-infringement-lawsuit, 2025, [Online; accessed 13-June-2025]

2024
[11]

Pelly, Mood Machine: The Rise of Spotify and the Costs of the Perfect Playlist

L. Pelly, Mood Machine: The Rise of Spotify and the Costs of the Perfect Playlist . Hodder & Stoughton, 2025

2025
[12]

Exploring the Use of Abusive Generative AI Models on Civitai,

Y . Wei, Y . Zhu, P. Hui, and G. Tyson, “Exploring the Use of Abusive Generative AI Models on Civitai,” in Proceedings of the 32nd ACM International Confer- ence on Multimedia, 2024, pp. 6949–6958

2024
[13]

AI Art and its Impact on Artists,

H. H. Jiang, L. Brown, J. Cheng, M. Khan, A. Gupta, D. Workman, A. Hanna, J. Flowers, and T. Gebru, “AI Art and its Impact on Artists,” in AIES. ACM, 2023

2023
[14]

From Melting Pots to Misrepresentations: Exploring Harms in Gen- erative AI,

S. Gautam, P. N. Venkit, and S. Ghosh, “From Melting Pots to Misrepresentations: Exploring Harms in Gen- erative AI,” in GenAICHI, 2024

2024
[15]

Slopa- ganda: The interaction between propaganda and gener- ative AI,

M. Klincewicz, M. Alfano, and A. E. Fard, “Slopa- ganda: The interaction between propaganda and gener- ative AI,” Filosofiska Notiser, vol. 12, no. 1, pp. 135– 162, 2025

2025
[18]

Study on the economic impact of Gen- erative AI in the Music and Audiovisual indus- tries,

CISAC, “Study on the economic impact of Gen- erative AI in the Music and Audiovisual indus- tries,” https://www.cisac.org/Newsroom/news- releases/global-economic-study-shows-human- creators-future-risk-generative-ai, 2024, [Online; accessed 22-March-2025]

2024
[19]

From Audio Deepfake Detection to AI-Generated Music Detection–A Pathway and Overview,

Y . Li, M. Milling, L. Specia, and B. W. Schuller, “From Audio Deepfake Detection to AI-Generated Music Detection–A Pathway and Overview,” arXiv preprint arXiv:2412.00571, 2024

2024 arXiv
[20]

Deconvo- lution and Checkerboard Artifacts,

A. Odena, V . Dumoulin, and C. Olah, “Deconvo- lution and Checkerboard Artifacts,” Distill, 2016. [Online]. Available: http://distill.pub/2016/deconv- checkerboard

2016
[21]

SONICS: Synthetic Or Not - Identi- fying Counterfeit Songs,

M. A. Rahman, Z. I. A. Hakim, N. H. Sarker, B. Paul, and S. A. Fattah, “SONICS: Synthetic Or Not - Identi- fying Counterfeit Songs,” in International Conference on Learning Representations (ICLR), 2025

2025
[22]

AI-Generated Music Detection and its Challenges,

D. Afchar, G. Meseguer-Brocal, and R. Hennequin, “AI-Generated Music Detection and its Challenges,” in ICASSP. IEEE, 2025

2025
[23]

ASVspoof: the automatic speaker verification spoofing and countermeasures challenge,

Z. Wu, J. Yamagishi, T. Kinnunen, C. Hanilçi, M. Sahidullah, A. Sizov, N. Evans, M. Todisco, and H. Delgado, “ASVspoof: the automatic speaker verification spoofing and countermeasures challenge,” IEEE Journal of Selected Topics in Signal Processing, vol. 11, no. 4, 2017

2017
[24]

A review of modern audio deepfake detection methods: challenges and fu- ture directions,

Z. Almutairi and H. Elgibreen, “A review of modern audio deepfake detection methods: challenges and fu- ture directions,” Algorithms, vol. 15, no. 5, p. 155, 2022

2022
[25]

AI-synthesized voice detection using neural vocoder artifacts,

C. Sun, S. Jia, S. Hou, and S. Lyu, “AI-synthesized voice detection using neural vocoder artifacts,” in Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 904–912

2023
[26]

Singfake: Singing voice deepfake detection,

Y . Zang, Y . Zhang, M. Heydari, and Z. Duan, “Singfake: Singing voice deepfake detection,” in ICASSP. IEEE, 2024

2024
[27]

The creation and detection of deepfakes: A survey,

Y . Mirsky and W. Lee, “The creation and detection of deepfakes: A survey,” ACM computing surveys (CSUR), vol. 54, no. 1, 2021

2021
[28]

Detecting Multimedia Generated by Large AI Models: A Sur- vey,

L. Lin, N. Gupta, Y . Zhang, H. Ren, C.-H. Liu, F. Ding, X. Wang, X. Li, L. Verdoliva, and S. Hu, “Detecting Multimedia Generated by Large AI Models: A Sur- vey,”arXiv:2402.00045, 2024

2024 arXiv
[29]

Detecting and simulating artifacts in gan fake images,

X. Zhang, S. Karaman, and S.-F. Chang, “Detecting and simulating artifacts in gan fake images,” in 2019 IEEE international workshop on information forensics and security (WIFS). IEEE, 2019, pp. 1–6

2019
[30]

CycleGAN without checkerboard artifacts for counter-forensics of fake-image detection,

T. Osakabe, M. Tanaka, Y . Kinoshita, and H. Kiya, “CycleGAN without checkerboard artifacts for counter-forensics of fake-image detection,” in Inter- national Workshop on Advanced Imaging Technology (IWAIT) 2021, vol. 11766. SPIE, 2021, pp. 51–55

2021
[31]

Detecting generated images by real images,

B. Liu, F. Yang, X. Bi, B. Xiao, W. Li, and X. Gao, “Detecting generated images by real images,” in Euro- pean Conference on Computer Vision. Springer, 2022, pp. 95–110

2022
[32]

Intriguing properties of synthetic im- ages: from generative adversarial networks to diffusion models,

R. Corvi, D. Cozzolino, G. Poggi, K. Nagano, and L. Verdoliva, “Intriguing properties of synthetic im- ages: from generative adversarial networks to diffusion models,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2023, pp. 973–982

2023
[33]

Adver- sarial audio synthesis,

C. Donahue, J. McAuley, and M. Puckette, “Adver- sarial audio synthesis,” in International Conference on Learning Representations, 2018

2018
[34]

Melgan: Generative adversarial networks for conditional waveform synthesis,

K. Kumar, R. Kumar, T. De Boissiere, L. Gestin, W. Z. Teoh, J. Sotelo, A. De Brebisson, Y . Bengio, and A. C. Courville, “Melgan: Generative adversarial networks for conditional waveform synthesis,”Advances in neu- ral information processing systems, vol. 32, 2019

2019
[35]

Upsam- pling artifacts in neural audio synthesis,

J. Pons, S. Pascual, G. Cengarle, and J. Serrà, “Upsam- pling artifacts in neural audio synthesis,” in ICASSP. IEEE, 2021

2021
[36]

Mallat, A Wavelet Tour of Signal Processing, Third Edition: The Sparse Way , 3rd ed

S. Mallat, A Wavelet Tour of Signal Processing, Third Edition: The Sparse Way , 3rd ed. USA: Academic Press, Inc., 2008

2008
[37]

A guide to convolu- tion arithmetic for deep learning,

V . Dumoulin and F. Visin, “A guide to convolu- tion arithmetic for deep learning,” arXiv preprint arXiv:1603.07285, 2016

2016 arXiv
[38]

High Fidelity Neural Audio Compression,

A. D’efossez, J. Copet, G. Synnaeve, and Y . Adi, “High Fidelity Neural Audio Compression,” ArXiv, vol. abs/2210.13438, 2022. [Online]. Available: https: //api.semanticscholar.org/CorpusID:253097788

2022 arXiv
[39]

High-Fidelity Audio Compression with Improved RVQGAN,

R. Kumar, P. Seetharaman, A. Luebs, I. Kumar, and K. Kumar, “High-Fidelity Audio Compression with Improved RVQGAN,”NeurIPS, vol. 36, 2024

2024
[40]

Musika! Fast Infinite Wave- form Music Generation,

M. Pasini and J. Schlüter, “Musika! Fast Infinite Wave- form Music Generation,” in ISMIR, 2022

2022
[41]

FMA: A dataset for music analysis,

M. Defferrard, K. Benzi, P. Vandergheynst, and X. Bresson, “FMA: A dataset for music analysis,” in ISMIR, 2017. [Online]. Available: https://arxiv.org/ abs/1612.01840

2017 arXiv
[42]

VampNet: Music Generation via Masked Acoustic Token Modeling,

H. F. Garcia, P. Seetharaman, R. Kumar, and B. Pardo, “VampNet: Music Generation via Masked Acoustic Token Modeling,”ISMIR, 2023

2023
[43]

Simple and Control- lable Music Generation,

J. Copet, F. Kreuk, I. Gat, T. Remez, D. Kant, G. Syn- naeve, Y . Adi, and A. Défossez, “Simple and Control- lable Music Generation,” NeurIPS, vol. 36, 2024

2024
[44]

Evaluation of Algorithms Using Games: The Case of Music Tagging,

E. Law, K. West, M. I. Mandel, M. Bay, and J. S. Downie, “Evaluation of Algorithms Using Games: The Case of Music Tagging,” inISMIR. Citeseer, 2009, pp. 387–392

2009
[45]

The MTG-Jamendo Dataset for Automatic Music Tagging,

D. Bogdanov, M. Won, P. Tovstogan, A. Porter, and X. Serra, “The MTG-Jamendo Dataset for Automatic Music Tagging,” in Machine Learning for Music Discovery Workshop, International Conference on Machine Learning (ICML 2019) , Long Beach, CA, United States, 2019. [Online]. Availa...

2019

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.

[1] [16]

Provocations from the Humanities for Generative AI Research,

L. Klein, M. Martin, A. Brock, M. Antoniak, M. Walsh, J. M. Johnson, L. Tilton, and D. Mimno, “Provocations from the Humanities for Generative AI Research,” arXiv preprint arXiv:2502.19190, 2025

arXiv 2025

[2] [17]

AI Art is Theft: Labour, Extraction, and Exploitation: Or, On the Dangers of Stochastic Pol- locks,

T. S. Goetze, “AI Art is Theft: Labour, Extraction, and Exploitation: Or, On the Dangers of Stochastic Pol- locks,” in ACM FAccT, 2024

work page 2024

[3] [1]

It’s not really enjoyable to make music now [...] I think the majority of people don’t enjoy the majority of the time they spend making music

INTRODUCTION “It’s not really enjoyable to make music now [...] I think the majority of people don’t enjoy the majority of the time they spend making music. ”— M. Shulman, CEO at Suno. Meanwhile, millions of users seem to enjoy creating AI-generated music. As a result, it was recently reported that at least a fifth of music delivered to streaming plat- fo...

work page 2025

[4] [2]

pitched noise

RELATED WORK 2.1 AI-music detection Mirroring the recent boom of commercial AI-music gener- ation services (e. g.,Suno, Udio [16]), the task of AI-music detection is relatively novel. Only a few works have been published so far [14, 16, 17]. These early works propose CNN-like models to learn to classify real and synthetic sig- nals. They discuss several c...

arXiv 2025

[5] [3]

de- convolution

FOURIER ANALYSIS OF ARTIFACTS In this work, we propose to reinterpret Convolutional Neu- ral Networks ( CNN) under the lenses of Fourier trans- forms. Instead of viewing inputs and outputs and hidden layers as time-based signals, we analyze their frequency- based decompositions (i. e., spectra). In the field of MIR, it is frequent to use spectrograms to p...

work page

[6] [4]

EXPERIMENT In this section, we validate our theoretical findings and close the gaps left out by theory with empirical results. Our research questions are the following: RQ1 Are the artifacts solely architecture dependent? RQ2 Can they be used to detect synthetic music? We show that the models we study (both open-source and closed-source) exhibit peaks in ...

work page

[7] [5]

CONCLUSION We propose a theoretical analysis for the emergence of peak artifacts in AI-music, formalize their recursive shape, and predict their architecture independence. Our exper- iments confirmed this latter observation, and we have used our new-found knowledge to craft a simple detector with performances on par with previous million-parameter models....

work page

[8] [6]

Deezer deploys cutting-edge AI detec- tion tool for music streaming,

Deezer, “Deezer deploys cutting-edge AI detec- tion tool for music streaming,” https://newsroom- deezer.com/2025/04/deezer-reveals-18-of-all-new- music-uploaded-to-streaming-is-fully-ai-generated/, 2025, [Online; accessed 22-March-2025]

work page 2025