Pith. sign in

REVIEW 3 major objections 5 minor 42 references

MusicMark embeds multi-bit watermarks into music’s semantic latents during diffusion so they ride with the music itself and survive codecs, cuts, and cover-song voice swaps without spoiling quality.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-14 06:56 UTC pith:BGW7QE2W

load-bearing objection Solid first generative music watermarking system with real codec/cover-song gains over post-hoc baselines; the semantic-latent story is useful but only partly stress-tested because training and eval share the same attack families. the 3 major comments →

arxiv 2607.11117 v1 pith:BGW7QE2W submitted 2026-07-13 cs.SD cs.AIcs.CR

MusicMark: A Robust Generative Watermarking Framework for Music Generation

Seohwan Yun , Jeeyoung Yun , Yongjin Kim , Juyeon Lee , Sungwoong Kim This is my paper
classification cs.SD cs.AIcs.CR
keywords music watermarkinggenerative watermarkingsemantic latent embeddingdiffusion music generationneural codec robustnesscover-song attackprovenance verification
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Commercial AI music is flooding the web, but existing audio watermarks were built for speech and are usually painted on after generation as faint noise. That residual signal is easy to strip—especially with modern neural codecs that keep the song’s meaning and throw away everything else—and the watermark step can simply be skipped. MusicMark claims a different route: inject the message into the semantic latent while the diffusion model is still denoising, so the watermark becomes part of the musical content rather than a post-production overlay. A lightweight adapter and detector are trained on a frozen generator with a joint objective that keeps watermarked latents close to their clean twins, plus heavy attack training that includes codec re-synthesis and a new cover-song pipeline. If the claim holds, provenance for generated music can no longer be bypassed by re-encoding or re-voicing, while the songs still sound like the original model’s output.

Core claim

MusicMark is presented as the first generative watermarking framework for lyrics- and text-conditioned music: by conditioning a diffusion latent denoiser on multi-bit messages through a decoupled watermark cross-attention adapter, the watermark is written into the semantic representation itself. Joint training with latent-consistency, fidelity, and watermark losses plus attack augmentations yields far higher detection and message-recovery rates than post-hoc baselines under neural codecs and cover-song transformations, while objective and human quality scores stay comparable to the unwatermarked backbone.

What carries the argument

The watermark adapter: a zero-initialized, decoupled cross-attention branch that injects a projected multi-bit message embedding into selected late layers of a frozen diffusion music model, combined with a latent consistency loss that keeps watermarked latents near their unwatermarked references.

Load-bearing premise

That the strong robustness numbers will still hold when real attackers use codecs, voice converters, or removal methods that were never seen in the training attack pool.

What would settle it

Take a held-out set of MusicMark tracks, re-encode them with a neural codec or voice-conversion stack never used in training or evaluation, then measure whether absolute message accuracy collapses toward the post-hoc baseline levels reported in Table I.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. MusicMark proposes the first generative watermarking framework for lyrics- and text-conditioned music generation. It freezes a diffusion backbone (primarily ACE-Step) and inserts a lightweight watermark adapter that embeds multi-bit messages into the semantic latent via a decoupled cross-attention branch during denoising, then trains a detector jointly with generation, fidelity, and watermark losses, including a latent consistency regularizer that keeps watermarked latents close to stop-gradient unwatermarked references. Attack augmentations (including neural codecs and a cover-song pipeline) are used during training. Empirically, MusicMark substantially outperforms post-hoc baselines (WavMark, AudioSeal, AudioSeal-M trained on the same data/attacks, SilentCipher) on detection and message extraction under 20 attacks—especially neural codec re-synthesis and cover-song + cut—while preserving objective quality (FAD/CLAP/PER/aesthetics) and human MOS relative to the unwatermarked backbone (Tables I–III), with ablations on stage, injection method, layer position, latent loss, and transfer to Stable Audio 3 (Tables IV–V).

Significance. If the results hold under broader evaluation, this is a solid systems contribution for provenance of AI-generated music: generative semantic-latent watermarking is a natural response to the known fragility of post-hoc audio watermarks under neural codecs, and the cover-song attack is a useful music-specific stress test. Strengths include a fair same-data/same-attack re-training of AudioSeal-M, multi-metric quality evaluation with human MOS, systematic ablations (stage, shared vs. decoupled attention, layer position, latent loss), and a backbone-transfer experiment on SA3. The work is timely given commercial AI music platforms and the documented failure modes of residual-signal watermarks under codec re-synthesis.

major comments (3)
  1. §III-F, Appendix C, Table VII, and Table I: robustness is optimized under the same attack families later reported as wins (EnCodec/DAC/SNAC, MP3, cover-song RVC, cut, speed, etc.). AudioSeal-M matches the training pool, which helps isolate the embedding locus, but absolute accuracy still collapses under DAC/SNAC (0.342/0.261) and Cover Song + Cut (0.813), and no held-out codec family, alternative voice-conversion stack, or detector-aware removal is reported. The central claim that semantic-latent embedding is inherently more robust than post-hoc residual insertion is only partially stress-tested; at least one truly unseen transform suite (or adaptive attack) is needed to support the interpretation beyond matched-augmentation superiority.
  2. §IV-B and Table I (Neural Codec / Cover Song rows): the paper’s strongest narrative is robustness under neural codecs and music-specific edits, yet Abs under DAC and SNAC remains low (0.342 and 0.261) despite perfect Acc and high Bit. The abstract and conclusion should more carefully distinguish detection reliability from full-message recovery under the hardest codecs, and clarify what message capacity / payload reliability is claimed for practical provenance use.
  3. §V and experimental scope: message capacity is fixed at N=16 bits and evaluation is mainly text/lyrics-conditioned generation (plus one SA3 style-only transfer). For a provenance framework, the manuscript should either demonstrate higher-capacity settings or more explicitly bound the claim (e.g., short identifiers only) and discuss collision / multi-user attribution limits under that capacity.
minor comments (5)
  1. SilentCipher Acc is marked “–” in Table I because it lacks a detection head; state this once in the table caption and avoid implying detection parity in prose summaries of “all metrics.”
  2. Fig. 3 difference visualizations are informative but would benefit from a fixed color scale / gain normalization so amplitude attenuation (SilentCipher) and band artifacts (AudioSeal-M) are comparable across columns.
  3. Notation: α is both the learnable watermark scale (Eq. 7) and appears as a training “Scale Factor α=0.1” in Table VI; disambiguate initialization vs. learned scale.
  4. Appendix B filtering (style–music similarity ≥0.5, English-only Muse clips) should note possible domain shift relative to the Suno-metadata evaluation prompts.
  5. Straight-through estimator for non-differentiable attacks is stated but not analyzed; a short note on whether STE bias affects codec vs. cover-song gradients would help reproducibility.

Circularity Check

0 steps flagged

No derivation circularity: empirical systems paper with independent baselines and ablations; train–eval attack overlap is a generalization limit, not a by-construction reduction.

full rationale

MusicMark is an empirical generative-watermarking systems paper, not a first-principles derivation that equates a claimed prediction to a fitted constant. The load-bearing claims are experimental comparisons (Tables I–III, IV–V): a frozen diffusion backbone plus a trainable decoupled watermark cross-attention adapter and detector, trained with a joint generation/fidelity/watermark objective (Eqs. 11–23), outperforms post-hoc baselines under listed attacks while preserving quality. Latent consistency (Eq. 14) is an explicit regularizer balanced against watermark loss, not a tautology that forces the reported Acc/Bit/Abs. AudioSeal-M is trained on the same data and attack pool, so the residual gap is not forced by unmatched training. There is no self-definitional equation, no uniqueness theorem imported from the same authors, no ansatz smuggled via self-citation, and no renaming of a known closed-form result. Overlap between training augmentations (§III-F, Appendix C, Table VII) and evaluation attacks is a standard generalization caveat (unseen codecs/adaptive removal untested), not circularity of the kind where a prediction reduces to its inputs by construction. Score 0; steps empty.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 4 invented entities

The central empirical claim rests on standard diffusion/flow-matching generation, frozen ACE-Step (and SA3) backbones, a hand-designed adapter/detector, and many training knobs (loss weights, adapter depth, α, 16-bit payload, attack schedules). No new physical entities are postulated; the “invented” pieces are engineering modules whose only evidence is in-paper metrics. Free parameters are the usual ML hyperparameter surface that shapes the robustness–quality trade-off.

free parameters (6)
  • Loss weights (λ_flow, λ_latent, λ_det, λ_msg, λ_1, λ_msspec, λ_msstft)
    Hand-set coefficients in Table VI that balance generation, latent consistency, fidelity, and watermark objectives; the reported trade-off depends on these choices.
  • Watermark scale α and zero-init K/V projections
    Learnable scale initialized at 0.1 with zero-init watermark keys/values controls how strongly the message enters latents without a uniqueness derivation.
  • Adapter layer placement (last 6 of 24 layers)
    Architectural choice validated by ablation; mid-layer variants underperform, so the main result depends on this placement.
  • Message length N=16 bits
    Fixed payload capacity used throughout training/eval; limits Abs interpretation and is acknowledged as a limitation.
  • Attack augmentation schedule and parameter ranges
    Attacks start after 8K steps; SNR/cutoff/bitrate/codebook ranges (Table VII) are chosen by authors and strongly shape measured robustness.
  • Learning rate schedule and training length (1e-4→1e-5, 60K steps, batch 8)
    Optimization hyperparameters that determine whether the joint objective converges to the reported operating point.
axioms (5)
  • domain assumption Diffusion/flow-matching latent music generators produce a semantic latent that neural codecs and cover-song pipelines largely preserve while discarding residual post-hoc signals.
    Motivates semantic-latent embedding over post-hoc perturbations (§I, neural codec discussion).
  • domain assumption Freezing the base generator and training only adapter+detector preserves base quality if watermarked latents stay close to unwatermarked references.
    Underpins L_latent and the frozen-backbone design (§III-C, III-F).
  • ad hoc to paper Straight-through estimator is an acceptable surrogate for non-differentiable attacks (MP3, codecs, cover-song) during joint training.
    Used so detector gradients ignore discrete attack implementations (§III-F).
  • domain assumption Standard signal-processing and codec attack models plus one RVC-based cover-song pipeline are representative enough to claim robustness for provenance use.
    Evaluation design in §IV-A and Appendix C; not proven against adaptive attackers.
  • standard math Flow matching / attention algebra and cross-entropy bit losses are valid training objectives.
    Standard ML optimization machinery in §III-F.
invented entities (4)
  • MusicMark watermark adapter (bit embedding table + decoupled watermark cross-attention) no independent evidence
    purpose: Inject multi-bit messages into semantic latents across denoising steps without rewriting text conditioning.
    Core module; evidence is only in-paper robustness/quality metrics, not external independent measurement.
  • Latent consistency loss L_latent between watermarked and stop-grad unwatermarked denoised latents no independent evidence
    purpose: Regularize fidelity so watermarking does not drift content latents.
    Paper claims first such regularizer for generative music watermarking; support is ablation Table IV(e).
  • Cover-song attack evaluation pipeline (MDX-Net separation + RVC vocal conversion + optional cut) no independent evidence
    purpose: Music-specific stress test preserving melody/harmony/instrumentals while changing singer identity.
    Useful benchmark construct; not an external standard with independent calibration beyond this paper.
  • AudioSeal-M (44.1 kHz stereo re-train of AudioSeal on same data/attacks) no independent evidence
    purpose: Fairer post-hoc baseline matched to MusicMark’s training regime.
    Author-constructed baseline; necessary for comparison but not independently published as a community artifact here.

pith-pipeline@v1.1.0-grok45 · 24311 in / 4120 out tokens · 40403 ms · 2026-07-14T06:56:02.123224+00:00 · methodology

0 comments
read the original abstract

AI music generation has rapidly advanced alongside commercial platforms, raising the need for reliable watermarking for provenance and attribution. However, existing audio watermarking research has largely focused on speech, and applying speech-oriented methods to music is challenging due to music's complex structure and rich acoustic texture. Most existing methods are post-hoc, adding imperceptible perturbations after generation rather than embedding watermarks as part of the content. This makes them fragile under transformations and especially vulnerable to neural codec re-synthesis, which can discard imperceptible residual signals. Moreover, since generation and watermarking are decoupled, the watermarking step can be bypassed or omitted, weakening provenance guarantees. To address these issues, we propose MusicMark, which, to the best of our knowledge, is the first generative watermarking framework for music. Specifically, MusicMark embeds watermark messages into the semantic latent space during generation, incorporating the watermark as part of the musical content and ensuring robustness against diverse attacks, particularly neural codec re-synthesis. To this end, we introduce a watermark adapter into a diffusion-based generation model to embed watermark messages across denoising steps. The adapter and detector are trained with a joint objective that preserves fidelity by constraining watermarked latents close to their unwatermarked reference latents, while improving robustness through attack augmentations. Experiments demonstrate that MusicMark substantially outperforms post-hoc baselines across diverse attacks including neural codec re-synthesis, while maintaining comparable generation quality. We further introduce a cover-song attack, converting the singing voice while preserving musical content, and show that MusicMark remains more robust than post-hoc methods.

Figures

Figures reproduced from arXiv: 2607.11117 by Jeeyoung Yun, Juyeon Lee, Seohwan Yun, Sungwoong Kim, Yongjin Kim.

Figure 1
Figure 1. Figure 1: Overview of MusicMark. a) Inference process, where lyrics and style tags are used as text conditions, and the watermark adapter injects the watermark [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the cover-song attack pipeline. A watermarked music sample is separated into instrumental and vocal stems, and the vocal stem is [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative comparison of music watermarking methods. The Original column shows an unwatermarked audio excerpt, and the remaining columns [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

42 extracted references · 15 linked inside Pith

  1. [1]

    Levo: High-quality song generation with multi- preference alignment,

    S. Lei, Y . Xu, H. Zhang, H. Chen, Y . Zhang, C. Yang, H. Zhu, S. Wang, Z. Wu, D. Yuet al., “Levo: High-quality song generation with multi- preference alignment,”Advances in Neural Information Processing Systems, vol. 38, pp. 102 448–102 479, 2026. PREPRINT 11

    2026

  2. [2]

    Ace-step: A step towards music generation foundation model,

    J. Gong, S. Zhao, S. Wang, S. Xu, and J. Guo, “Ace-step: A step towards music generation foundation model,”arXiv preprint arXiv:2506.00045, 2025

    Pith/arXiv arXiv 2025

  3. [3]

    Diffrhythm: Blazingly fast and embarrassingly simple end-to- end full-length song generation with latent diffusion,

    Z. Ning, H. Chen, Y . Jiang, C. Hao, G. Ma, S. Wang, J. Yao, and L. Xie, “Diffrhythm: Blazingly fast and embarrassingly simple end-to- end full-length song generation with latent diffusion,”arXiv preprint arXiv:2503.01183, 2025

    Pith/arXiv arXiv 2025

  4. [4]

    Diffrhythm 2: Efficient and high fidelity song generation via block flow matching,

    Y . Jiang, H. Chen, Z. Ning, J. Yao, Z. Han, D. Wu, M. Meng, J. Luan, Z. Fu, and L. Xie, “Diffrhythm 2: Efficient and high fidelity song generation via block flow matching,”arXiv preprint arXiv:2510.22950, 2025

    arXiv 2025

  5. [5]

    Yue: Scaling open foundation models for long- form music generation,

    R. Yuan, H. Lin, S. Guo, G. Zhang, J. Pan, Y . Zang, H. Liu, Y . Liang, W. Ma, X. Duet al., “Yue: Scaling open foundation models for long- form music generation,”arXiv preprint arXiv:2503.08638, 2025

    arXiv 2025

  6. [6]

    Heartmula: A family of open sourced music foundation models,

    D. Yang, Y . Xie, Y . Yin, Z. Wang, X. Yi, G. Zhu, X. Weng, Z. Xiong, Y . Ma, D. Conget al., “Heartmula: A family of open sourced music foundation models,”arXiv preprint arXiv:2601.10547, 2026

    Pith/arXiv arXiv 2026

  7. [7]

    Jam: A tiny flow-based song generator with fine-grained controllability and aesthetic alignment,

    R. Liu, C.-Y . Hung, N. Majumder, T. Gautreaux, A. A. Bagherzadeh, C. Li, D. Herremans, and S. Poria, “Jam: A tiny flow-based song generator with fine-grained controllability and aesthetic alignment,” arXiv preprint arXiv:2507.20880, 2025

    Pith/arXiv arXiv 2025

  8. [8]

    Stable audio 3,

    Z. Evans, J. D. Parker, M. Rice, C. Carr, Z. Zukowski, J. Taylor, and J. Pons, “Stable audio 3,”arXiv preprint arXiv:2605.17991, 2026

    Pith/arXiv arXiv 2026

  9. [9]

    Suno AI Music Generator,

    Suno, Inc., “Suno AI Music Generator,” https://suno.com, 2026, ac- cessed: 2026-05-18

    2026

  10. [10]

    Udio AI Music Generator,

    Udio, “Udio AI Music Generator,” https://www.udio.com/, 2026, ac- cessed: 2026-05-18

    2026

  11. [11]

    Proactive detection of voice cloning with localized watermarking,

    R. San Roman, P. Fernandez, H. Elsahar, A. D´efossez, T. Furon, and T. Tran, “Proactive detection of voice cloning with localized watermarking,”ICML, 2024

    2024

  12. [12]

    Silentcipher: Deep audio watermarking

    M. K. Singh, N. Takahashi, W.-H. Liao, and Y . Mitsufuji, “Silentcipher: Deep audio watermarking.” inINTERSPEECH, 2024

    2024

  13. [13]

    Wavmark: Watermarking for audio generation,

    G. Chen, Y . Wu, S. Liu, T. Liu, X. Du, and F. Wei, “Wavmark: Watermarking for audio generation,”arXiv preprint arXiv:2308.12770, 2023

    Pith/arXiv arXiv 2023

  14. [14]

    Detecting voice cloning attacks via timbre watermarking,

    C. Liu, J. Zhang, T. Zhang, X. Yang, W. Zhang, and N. Yu, “Detecting voice cloning attacks via timbre watermarking,” inNetwork and Dis- tributed System Security Symposium, 2024

    2024

  15. [15]

    Dear: A deep-learning-based audio re-recording resilient watermarking,

    C. Liu, J. Zhang, H. Fang, Z. Ma, W. Zhang, and N. Yu, “Dear: A deep-learning-based audio re-recording resilient watermarking,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 37, no. 11, 2023, pp. 13 201–13 209

    2023

  16. [16]

    Traceablespeech: Towards proactively traceable text- to-speech with watermarking,

    J. Zhou, J. Yi, T. Wang, J. Tao, Y . Bai, C. Y . Zhang, Y . Ren, and Z. Wen, “Traceablespeech: Towards proactively traceable text- to-speech with watermarking,” 2024. [Online]. Available: https: //arxiv.org/abs/2406.04840

    Pith/arXiv arXiv 2024

  17. [17]

    Hifi-ganw: Watermarked speech synthesis via fine-tuning of hifi-gan,

    X. Cheng, Y . Wang, C. Liu, D. Hu, and Z. Su, “Hifi-ganw: Watermarked speech synthesis via fine-tuning of hifi-gan,”IEEE Signal Processing Letters, vol. 31, pp. 2440–2444, 2024

    2024

  18. [18]

    V ocbulwark: Towards practical generative speech watermarking via additional-parameter injection,

    W. Liu, Y . Li, and Z. Yin, “V ocbulwark: Towards practical generative speech watermarking via additional-parameter injection,”arXiv preprint arXiv:2601.22556, 2026

    arXiv 2026

  19. [19]

    Jen-1: Text-guided universal music generation with omnidirectional diffusion models,

    P. P. Li, B. Chen, Y . Yao, Y . Wang, A. Wang, and A. Wang, “Jen-1: Text-guided universal music generation with omnidirectional diffusion models,” in2024 IEEE Conference on Artificial Intelligence (CAI). IEEE, 2024, pp. 762–769

    2024

  20. [20]

    Musiclm: Generating music from text,

    A. Agostinelli, T. I. Denk, Z. Borsos, J. Engel, M. Verzetti, A. Caillon, Q. Huang, A. Jansen, A. Roberts, M. Tagliasacchiet al., “Musiclm: Generating music from text,”arXiv preprint arXiv:2301.11325, 2023

    Pith/arXiv arXiv 2023

  21. [21]

    Simple and controllable music generation,

    J. Copet, F. Kreuk, I. Gat, T. Remez, D. Kant, G. Synnaeve, Y . Adi, and A. D ´efossez, “Simple and controllable music generation,”Advances in neural information processing systems, vol. 36, pp. 47 704–47 720, 2023

    2023

  22. [22]

    Mert: Acoustic music understanding model with large-scale self-supervised training,

    Y . Li, R. Yuan, G. Zhang, Y . Ma, X. Chen, H. Yin, C. Xiao, C. Lin, A. Ragni, E. Benetoset al., “Mert: Acoustic music understanding model with large-scale self-supervised training,” inInternational Conference on Learning Representations, vol. 2024, 2024, pp. 12 181–12 204

    2024

  23. [23]

    High fidelity neural audio compression,

    A. D ´efossez, J. Copet, G. Synnaeve, and Y . Adi, “High fidelity neural audio compression,”arXiv preprint arXiv:2210.13438, 2022

    Pith/arXiv arXiv 2022

  24. [24]

    Snac: Multi-scale neural audio codec,

    H. Siuzdak, F. Gr ¨otschla, and L. A. Lanzend ¨orfer, “Snac: Multi-scale neural audio codec,” inAudio Imagination: NeurIPS 2024 Workshop AI-Driven Speech, Music, and Sound Generation, 2024

    2024

  25. [25]

    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,”Advances in Neural Information Processing Systems, vol. 36, pp. 27 980–27 993, 2023

    2023

  26. [26]

    A comprehensive real-world assessment of audio watermarking algo- rithms: Will they survive neural codecs?

    Y . ¨Ozer, W. Choi, J. Serr `a, M. K. Singh, W.-H. Liao, and Y . Mitsufuji, “A comprehensive real-world assessment of audio watermarking algo- rithms: Will they survive neural codecs?” inProceedings of the Annual Conference of the International Speech Communication Association (Interspeech), Rotterdam, The Netherlands, 2025

    2025

  27. [27]

    Sok: How robust is audio watermarking in generative ai models?

    Y . Wen, A. Innuganti, A. B. Ramos, H. Guo, and Q. Yan, “Sok: How robust is audio watermarking in generative ai models?”arXiv preprint arXiv:2503.19176, 2025

    Pith/arXiv arXiv 2025

  28. [28]

    Deep audio watermarks are shallow: Limitations of post-hoc watermarking techniques for speech,

    P. O’Reilly, Z. Jin, J. Su, and B. Pardo, “Deep audio watermarks are shallow: Limitations of post-hoc watermarking techniques for speech,” arXiv preprint arXiv:2504.10782, 2025

    Pith/arXiv arXiv 2025

  29. [29]

    Audiomarkbench: Benchmarking robustness of audio watermarking,

    H. Liu, M. Guo, Z. Jiang, L. Wang, and N. Z. Gong, “Audiomarkbench: Benchmarking robustness of audio watermarking,”Advances in Neural Information Processing Systems, vol. 37, pp. 52 241–52 265, 2024

    2024

  30. [30]

    Latent watermarking of audio generative models,

    R. San Roman, P. Fernandez, A. Deleforge, Y . Adi, and R. Serizel, “Latent watermarking of audio generative models,” inICASSP 2025- 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2025, pp. 1–5

    2025

  31. [31]

    Robust watermarks for audio diffu- sion models by quadrature amplitude modulation,

    K. Lee, S. Hong, and S. Y . Chun, “Robust watermarks for audio diffu- sion models by quadrature amplitude modulation,”Pattern Recognition Letters, 2025

    2025

  32. [32]

    Groot: Generating robust watermark for diffusion-model-based audio synthesis,

    W. Liu, Y . Li, D. Lin, H. Tian, and H. Li, “Groot: Generating robust watermark for diffusion-model-based audio synthesis,” inProceedings of the 32nd ACM International Conference on Multimedia, 2024, pp. 3294–3302

    2024

  33. [33]

    Robust and imperceptible watermarking framework for generative audio models,

    Y . Feng, X. Zhang, F. Feng, G. Zhang, and L. Xu, “Robust and imperceptible watermarking framework for generative audio models,” IEEE Signal Processing Letters, 2025

    2025

  34. [34]

    Songgen: A single stage auto-regressive transformer for text-to-song generation,

    Z. Liu, S. Ding, Z. Zhang, X. Dong, P. Zhang, Y . Zang, Y . Cao, D. Lin, and J. Wang, “Songgen: A single stage auto-regressive transformer for text-to-song generation,”arXiv preprint arXiv:2502.13128, 2025

    Pith/arXiv arXiv 2025

  35. [35]

    Understanding straight-through estimator in training activation quantized neural nets,

    P. Yin, J. Lyu, S. Zhang, S. Osher, Y . Qi, and J. Xin, “Understanding straight-through estimator in training activation quantized neural nets,” arXiv preprint arXiv:1903.05662, 2019

    Pith/arXiv arXiv 1903

  36. [36]

    Muse: Towards reproducible long- form song generation with fine-grained style control,

    C. Jiang, J. Chen, Z. Xiang, Z. Yang, H. Wang, J. Zhuang, X. Che, J. Sun, H. Li, Y . Caoet al., “Muse: Towards reproducible long- form song generation with fine-grained style control,”arXiv preprint arXiv:2601.03973, 2026

    arXiv 2026

  37. [37]

    Fr ´echet audio distance: A metric for evaluating music enhancement algorithms,

    K. Kilgour, M. Zuluaga, D. Roblek, and M. Sharifi, “Fr ´echet audio distance: A metric for evaluating music enhancement algorithms,”arXiv preprint arXiv:1812.08466, 2018

    Pith/arXiv arXiv 2018

  38. [38]

    Clap learning audio concepts from natural language supervision,

    B. Elizalde, S. Deshmukh, M. Al Ismail, and H. Wang, “Clap learning audio concepts from natural language supervision,” inICASSP 2023- 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2023, pp. 1–5

    2023

  39. [39]

    Meta audiobox aesthetics: Unified automatic quality assessment for speech, music, and sound,

    A. Tjandra, Y .-C. Wu, B. Guo, J. Hoffman, B. Ellis, A. Vyas, B. Shi, S. Chen, M. Le, N. Zacharovet al., “Meta audiobox aesthetics: Unified automatic quality assessment for speech, music, and sound,”arXiv preprint arXiv:2502.05139, 2025

    Pith/arXiv arXiv 2025

  40. [40]

    A binaural room impulse response database for the evaluation of dereverberation algorithms,

    M. Jeub, M. Schafer, and P. Vary, “A binaural room impulse response database for the evaluation of dereverberation algorithms,” in2009 16th international conference on digital signal processing. IEEE, 2009, pp. 1–5

    2009

  41. [41]

    Kuielab-mdx-net: A two-stream neural network for music demixing,

    M. Kim, W. Choi, J. Chung, D. Lee, and S. Jung, “Kuielab-mdx-net: A two-stream neural network for music demixing,” inMDX Workshop at ISMIR 2021, 2021

    2021

  42. [42]

    Robust speech recognition via large-scale weak supervi- sion,

    A. Radford, J. W. Kim, T. Xu, G. Brockman, C. McLeavey, and I. Sutskever, “Robust speech recognition via large-scale weak supervi- sion,” inInternational conference on machine learning. PMLR, 2023, pp. 28 492–28 518. APPENDIX A. Implementation Details MusicMark.MusicMark is implemented on ACE-Step [2], which consists of a Deep Compression AutoEncoder (DCA...

    2023