Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Fast Text-to-Audio Generation with Adversarial Post-Training

T0 review · 3 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 relativistic-contrastive post-training objective converts a pre-trained rectified flow text-to-audio model into an eight-step generator, matching distilled quality while preserving diversity and dropping classifier-free guidance.

desk verdict Promising method and strong experiments, but the core objective in Eq. (5)/(8) has an inverted sign that contradicts the prose; the paper is not reproducible as written. read the letter →

arxiv 2505.08175 v3 pith:ECQ5N7TP submitted 2025-05-13 cs.SD cs.AIcs.LGcs.MMeess.AS

Zachary Novack , Zach Evans , Zack Zukowski , Josiah Taylor , CJ Carr , Julian Parker , Adnan Al-Sinan , Gian Marco Iodice
show 3 more authors
Julian McAuley Taylor Berg-Kirkpatrick Jordi Pons
This is my paper · ORCID
classification cs.SDcs.AIcs.LGcs.MMeess.AS
keywords text-to-audiogenerationadversarialpost-trainingrectifiedflowrelativisticlosscontrastivediscriminatorclassifier-freeguidanceping-pongsamplingconditionaldiversity
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper introduces Adversarial Relativistic-Contrastive (ARC) post-training, an acceleration recipe for text-to-audio rectified flow models that needs neither distillation nor classifier-free guidance. It replaces the flow model's velocity-matching loss with a relativistic adversarial loss, paired with a contrastive discriminator objective, and claims this turns a pre-trained model into an eight-step generator competitive with distillation-based accelerators. The speed payoff is large: roughly twelve seconds of 44.1 kHz stereo audio in about 75 ms on an H100 GPU and about seven seconds on a phone CPU. The authors argue this makes interactive and on-device text-to-audio realistic without the training cost, memory overhead, or diversity loss of distillation.

What carries the argument

The load-bearing object is the ARC objective: a relativistic adversarial game over same-prompt real/generated audio pairs, defined in the paper as $\min_\phi \max_\psi \left[L_{\mathrm{R}}(\phi,\psi) + \lambda L_{\mathrm{C}}(\psi)\right]$. $L_{\mathrm{R}}$ is a relativistic logistic loss that rewards the generator for producing samples judged more real than their paired real sample, and $L_{\mathrm{C}}$ is a contrastive loss that trains only the discriminator to separate correct from shuffled audio-prompt pairs. The claimed interaction between the two, with the discriminator also initialized from the pre-trained flow model, is what substitutes for distillation and classifier-free guidance. Ping-pong sampling carries the final eight-step trajectory by alternating denoising and re-noising at decreasing noise levels.

What would settle it

Run ARC with and without the contrastive discriminator loss and compare eight-step prompt-adherence scores on the same prompts; if removing the contrastive loss changes nothing, the claimed semantic transfer from discriminator to generator is not happening. A second check is to test the trained discriminator alone on shuffled-prompt detection: if it cannot rank correct above incorrect prompts, the contrastive training did not make it semantically aware.

Watch

Extended reading notes

Core claim

ARC post-training is a min-max objective $L_{\mathrm{R}} + \lambda L_{\mathrm{C}}$ in which a text-conditioned discriminator compares pairs of noised real audio and noised generated audio that share the same prompt. The relativistic loss $L_{\mathrm{R}}$ pushes each generated sample to score as more real than its paired real sample, while the contrastive loss $L_{\mathrm{C}}$ additionally trains the same discriminator to score correct audio-prompt pairs above shuffled ones. The paper argues that this semantically aware discriminator makes the relativistic signal carry prompt adherence, so classifier-free guidance can be dropped; with ping-pong sampling the post-trained generator runs in eight steps. Across objective metrics and listening tests, the result is competitive with an eight-step distillation baseline while retaining substantially more generative diversity, and it is roughly one hundred times faster than the original hundred-step model.

Load-bearing premise

The load-bearing premise is that making the discriminator good at telling correct from shuffled audio-prompt pairs also makes the generator follow prompts better, even though the generator is never trained on that pairing signal directly.

Editorial extensions

If this is right

  • A pre-trained rectified flow text-to-audio model can be accelerated to eight sampling steps without distillation, avoiding the need to generate and store teacher trajectory-output pairs.
  • Because classifier-free guidance is removed, inference needs only a single text-conditioned forward pass per step, which cuts peak memory and is what makes CPU-based edge-device deployment practical.
  • Diversity is preserved better than with distillation-based acceleration, as measured by the paper's new same-prompt conditional diversity score and by listening tests.
  • The reported timings, about 75 ms for roughly twelve seconds of 44.1 kHz stereo audio on an H100, put text-to-audio latency in a range suitable for interactive creative tools.

Reading between the lines

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

  • Nothing in the ARC objective is audio-specific, so the same recipe should transfer to other conditional flow and diffusion models, such as image or video generators; that transfer is an extension the paper does not claim.
  • The paper does not train the generator with the contrastive loss, so a natural untested modification is adding a small contrastive term to the generator's update and checking whether prompt adherence improves further without eroding diversity.
  • The paper reports audio-to-audio style transfer by seeding ping-pong sampling with a recording; a testable extension would be systematic evaluation of this as a beat-alignment or voice-control mechanism.
  • The same-prompt diversity metric introduced here could serve as a standard evaluation companion for any conditional generative model that reports both fidelity and diversity.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper introduces ARC (Adversarial Relativistic-Contrastive) post-training, an acceleration method for pre-trained rectified-flow text-to-audio models. ARC replaces the velocity-matching loss with a relativistic adversarial loss computed on paired real and generated audio (Eq. 5) and adds a contrastive discriminator loss over correct versus shuffled prompt-audio pairs (Eq. 8). The generator is trained without distillation or classifier-free guidance and is sampled with a ping-pong scheme. On a variant of Stable Audio Open, the authors report 8-step generation of about 12 seconds of 44.1 kHz stereo audio in about 75 ms on an H100, about 187 ms on a 3090, and about 6.6 seconds on a mobile phone after quantization. Experiments on AudioCaps include objective metrics, a proposed conditional diversity metric (CCDS), ablations against Presto and standard least-squares adversarial losses, and a small listening test.

Significance. If the paper's equations and results are taken at face value, ARC is a potentially useful contribution: it is a non-distillation, CFG-free adversarial post-training recipe for audio that appears to preserve diversity better than the distillation baseline while matching its speed, and the on-device deployment experiments are a valuable practical data point. The paper also commits to public code and a demo, and the empirical evaluation uses an external benchmark and listening tests rather than fitted constants. The main caveat is that the two defining losses are written with reversed signs, so the method as specified is not reproducible; this must be corrected before the contribution can be assessed. The concern that the contrastive loss only trains the discriminator and may not propagate to the generator is partially addressed empirically by the LR vs. LR+LC ablation, which shows improved CLAP score and prompt adherence when LC is included, although no theoretical analysis is provided.

major comments (3)
  1. [Sec. 2.3, Eqs. (4)-(5)] The relativistic loss is sign-reversed. Since f(x) = -log(1+e^{-x}) is strictly increasing, maximizing LR with respect to D_psi pushes Delta_gen above Delta_real, and minimizing LR with respect to G_phi pushes Delta_gen below Delta_real. This is the opposite of the training goal stated in Sec. 2.3, where G should make generated samples 'more real' than their paired real sample and D should make real samples 'more real' than their paired generated sample. As written, the discriminator is rewarded for assigning higher logits to generated samples and the generator is rewarded for lowering them, which would not produce the reported quality. Please correct the sign (e.g., use f(x) = -log(1+e^{x}) or swap the arguments) and confirm that the text, equations, and released code agree; the reward-model interpretation in Sec. 2.6 is also reversed under the current sign.
  2. [Sec. 2.4, Eq. (8)] The contrastive loss is also reversed. With f increasing, maximizing LC trains D_psi to assign higher logits to shuffled incorrect prompt-audio pairs than to correct pairs, which is the opposite of a contrastive objective intended to make the discriminator reward correct audio-text alignment. The equation should use f(Delta_real(x0,s,c) - Delta_real(x0,s,P[c])) or a decreasing f; otherwise the method as written actively rewards wrong prompts and would be expected to degrade prompt adherence rather than improve it.
  3. [Abstract, Sec. 1, and Table 1] The speed claims are not supported by the paper's own table. In Table 1, +Presto and +ARC (ours) have identical RTF (156.42) at 8 steps, so 'the fastest text-to-audio model to our knowledge' is contradicted by the presented baseline, and no comparison is made to other cited fast text-to-audio models (e.g., the consistency-distillation models in refs. [26]-[29]). Additionally, the abstract and introduction claim the model is '100x faster' than SAO, but the Table 1 RTF values imply about a 44x latency improvement over 100-step SAO (3.56 vs. 156.42 in RTF, i.e., 12/3.56 s vs. 12/156.42 s). Please restrict the speed claims to the specific setting that supports them or add the missing comparisons.
minor comments (5)
  1. [Sec. 3.3 heading] There is a typo in the section title: 'Objective Evalaution' should be 'Objective Evaluation'.
  2. [Table 1] The objective metrics are reported without confidence intervals, error bars, or significance tests; adding variance across evaluation seeds or bootstrap intervals would strengthen the comparison, especially because several values are close (e.g., CLAP 0.27 for ARC and Presto).
  3. [Sec. 3.4] The listening test uses only 14 participants; please report the number of stimuli per condition and the statistical test used for the claims that differences are or are not significant, since the 95% confidence intervals alone do not establish significance across conditions.
  4. [Sec. 3.7] The sentence 'This decreases inference time from 15.3s (original F32) to 6.6s, and reducing peak runtime RAM usage...' has a grammatical issue ('and reducing' should be 'and reduces'); also clarify whether the reported edge times are for the same 12-second audio length and state the corresponding RTF on the device.
  5. [Sec. 3.5] The RTF column header would benefit from a footnote repeating the definition 'audio duration divided by latency' directly in the table caption, since the term is often defined in the opposite direction in the literature.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: ARC's central claims are empirical against external benchmarks and independent baselines.

full rationale

The paper's central claim is that the ARC objective in Eq. (4), a min-max of the relativistic loss L_R (Eq. 5) and the contrastive discriminator loss L_C (Eq. 8), accelerates a pre-trained rectified-flow text-to-audio model to 8-step generation without distillation or classifier-free guidance. The support is empirical: Table 1 reports external metrics (FD_openl3, KL_passt, CLAP, R_passt, C_passt), human MOS, RTF, and VRAM against independent baselines including the pre-trained RF, SAO, and Presto. No predicted quantity is equal to a fitted parameter by construction. L_C trains only the discriminator, and the generator is trained only through L_R, so the claim that a semantically aware D_psi improves prompt adherence via L_R is an empirical transfer verified by ablation (+L_R without L_C versus full ARC), not an identity. CLAP evaluation shares a general contrastive audio-text paradigm with L_C, but the generator is not trained directly against CLAP and the discriminator is not used at inference, so the metric is not forced. Self-citations (SAO, Presto, DITTO-2, and the Presto-style noise schedule) are used as base model, baseline, or training-detail context, not as load-bearing uniqueness theorems or hidden ansatze; none is imported to rule out alternatives. The sign inconsistency in Eq. (5)/(8) flagged by the skeptic is a correctness/reproducibility issue, not a circularity: if the equations are implemented literally, the min-max direction contradicts the prose, but that is a contradiction, not an equivalence between output and input. Therefore the derivation chain is self-contained and the circularity score is 0.

Assumptions & free parameters 8 free parameters · 6 assumptions · 0 invented entities

The central claim rests on a pre-trained flow model, a fixed autoencoder, chosen training hyperparameters, and the empirical validity of evaluation metrics. The most fragile item is the untested assumption that the contrastive discriminator objective transfers prompt-adherence signal to the generator through the relativistic loss.

free parameters (8)
  • lambda (contrastive loss weight) = 1
    Weight for LC in Eq. 4; chosen by hand, not swept.
  • Post-training iterations = 100k
    All acceleration methods are fine-tuned for 100k iterations; a fixed budget.
  • Learning rate = 5e-7
    Used for both generator and discriminator with AdamW; chosen by hand.
  • Batch size = 256
    Training batch size across 8 H100 GPUs; fixed.
  • Generator noise schedule pgen(t) = uniform in log-SNR from -6 to 2
    Noise distribution for generator training; chosen to align with inference.
  • Discriminator noise schedule pdisc(s) = shifted logit normal
    Noise distribution for discriminator inputs; chosen to emphasize mid-to-high SNR.
  • Number of sampling steps = 8
    Ping-pong steps used for the final model; ablation showed 8 steps work best.
  • Discriminator DiT block usage = 75%
    Initializes discriminator with the first 75% of the pre-trained DiT blocks; architecture choice.
assumptions (6)
  • domain assumption The pre-trained rectified flow model v_theta is an adequate starting point for adversarial post-training.
    Used in Sec. 2.2 and 3.1; the method fine-tunes this model rather than training from scratch.
  • domain assumption The SAO latent autoencoder faithfully encodes and decodes 44.1 kHz stereo audio.
    Sec. 3.1; the generator operates in this latent space and the autoencoder is held fixed.
  • domain assumption The contrastive loss on shuffled prompts teaches the discriminator semantic alignment that transfers to the generator through the relativistic loss.
    Sec. 2.4, Eqs. (4)-(8); this is the mechanism that replaces classifier-free guidance and is not formally proven.
  • domain assumption Objective metrics (FDopenl3, KLpasst, CLAP, Rpasst, Cpasst) and MOS are valid proxies for audio quality and prompt adherence.
    Sec. 3.3; all conclusions about quality and prompt adherence rely on these metrics.
  • domain assumption The AudioCaps test set is representative of general text-to-audio prompts.
    Sec. 3.3; all quantitative evaluation uses this single dataset.
  • domain assumption Ping-pong sampling is a valid inference procedure for a few-step generator trained to denoise.
    Sec. 2.5; used without an ablation against other sampling schedules.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Fast Text-to-Audio Generation with Adversarial Post-Training." pith.science (2026). https://pith.science/paper/ECQ5N7TP

@misc{pith2026250508175,
  author       = {Pith},
  title        = {Pith review of: Fast Text-to-Audio Generation with Adversarial Post-Training},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ECQ5N7TP}},
  note         = {Machine review of arXiv:2505.08175}
}
abstract

Text-to-audio systems, while increasingly performant, are slow at inference time, thus making their latency unpractical for many creative applications. We present Adversarial Relativistic-Contrastive (ARC) post-training, the first adversarial acceleration algorithm for diffusion/flow models not based on distillation. While past adversarial post-training methods have struggled to compare against their expensive distillation counterparts, ARC post-training is a simple procedure that (1) extends a recent relativistic adversarial formulation to diffusion/flow post-training and (2) combines it with a novel contrastive discriminator objective to encourage better prompt adherence. We pair ARC post-training with a number optimizations to Stable Audio Open and build a model capable of generating $\approx$12s of 44.1kHz stereo audio in $\approx$75ms on an H100, and $\approx$7s on a mobile edge-device, the fastest text-to-audio model to our knowledge.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. WildFX: A DAW-Powered Pipeline for In-the-Wild Audio FX Graph Modeling

    cs.SD 2025-07 conditional novelty 6.0 of 10

    WildFX generates multi-track audio datasets by rendering real DAW effect graphs with commercial plugins inside Docker, and demonstrates the pipeline on blind mixing-graph estimation.

Reference graph

Works this paper leans on

51 extracted references · 15 canonical work pages · cited by 1 Pith paper

  1. [26]

    Accelerating diffusion-based text-to-audio generation with consistency distillation,

    Y . Bai, T. Dang, D. Tran, K. Koishida, and S. Sojoudi, “Accelerating diffusion-based text-to-audio generation with consistency distillation,” in Interspeech, 2024

    work page 2024

  2. [29]

    DITTO-2: Distilled diffusion inference-time t-optimization for music generation,

    Z. Novack, J. McAuley, T. Berg-Kirkpatrick, and N. J. Bryan, “DITTO-2: Distilled diffusion inference-time t-optimization for music generation,” in ISMIR, 2024

    work page 2024

  3. [1]

    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,” in NeurIPS, 2023

    2023

  4. [2]

    Fast timing- conditioned latent audio diffusion,

    Z. Evans, C. Carr, J. Taylor, S. H. Hawley, and J. Pons, “Fast timing- conditioned latent audio diffusion,” in ICML, 2024

    work page 2024

  5. [3]

    Long- form music generation with latent diffusion,

    Z. Evans, J. Parker, C. Carr, Z. Zukowski, J. Taylor, and J. Pons, “Long- form music generation with latent diffusion,” arXiv:2404.10301, 2024

    arXiv 2024

  6. [4]

    Stable audio open,

    Z. Evans, J. D. Parker, C. Carr, Z. Zukowski, J. Taylor, and J. Pons, “Stable audio open,” arXiv:2407.14358, 2024

    arXiv 2024

  7. [5]

    Denoising diffusion implicit models,

    J. Song, C. Meng, and S. Ermon, “Denoising diffusion implicit models,” in ICLR, 2020

    work page 2020

  8. [6]

    Scaling rectified flow transformers for high-resolution image synthesis,

    P. Esser, S. Kulal, A. Blattmann, R. Entezari, J. M ¨uller, H. Saini, Y . Levi, D. Lorenz, A. Sauer, F. Boesel et al., “Scaling rectified flow transformers for high-resolution image synthesis,” in ICML, 2024

    work page 2024

Show all 51 references
  1. [7]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” in NeurIPS, 2020

    2020

  2. [8]

    One-step diffusion with distribution matching distillation,

    T. Yin, M. Gharbi, R. Zhang, E. Shechtman, F. Durand, W. T. Freeman, and T. Park, “One-step diffusion with distribution matching distillation,” arXiv:2311.18828, 2023

    2023 arXiv

  3. [9]

    Improved distribution matching distillation for fast image synthesis,

    T. Yin, M. Gharbi, T. Park, R. Zhang, E. Shechtman, F. Durand, and W. T. Freeman, “Improved distribution matching distillation for fast image synthesis,” arXiv:2405.14867, 2024

    2024 arXiv

  4. [10]

    Fast high-resolution image synthesis with latent adversarial diffusion distillation,

    A. Sauer, F. Boesel, T. Dockhorn, A. Blattmann, P. Esser, and R. Rombach, “Fast high-resolution image synthesis with latent adversarial diffusion distillation,” arXiv:2403.12015, 2024

    2024 arXiv

  5. [11]

    Adversarial diffusion distillation,

    A. Sauer, D. Lorenz, A. Blattmann, and R. Rombach, “Adversarial diffusion distillation,” arXiv:2311.17042, 2023

    2023 arXiv

  6. [12]

    Hyper-SD: Trajectory segmented consistency model for efficient image synthesis,

    Y . Ren, X. Xia, Y . Lu, J. Zhang, J. Wu, P. Xie, X. Wang, and X. Xiao, “Hyper-SD: Trajectory segmented consistency model for efficient image synthesis,” arXiv:2404.13686, 2024

    2024 arXiv

  7. [13]

    Phased consistency model,

    F.-Y . Wang, Z. Huang, A. W. Bergman, D. Shen, P. Gao, M. Lingelbach, K. Sun, W. Bian, G. Song, Y . Liu, H. Li, and X. Wang, “Phased consistency model,” arXiv:2405.18407, 2024

    2024 arXiv

  8. [14]

    Consistency models,

    Y . Song, P. Dhariwal, M. Chen, and I. Sutskever, “Consistency models,” in ICML, 2023

    2023

  9. [15]

    Simplifying, stabilizing and scaling continuous-time consistency models,

    C. Lu and Y . Song, “Simplifying, stabilizing and scaling continuous-time consistency models,” arXiv:2410.11081, 2024

    2024 arXiv

  10. [16]

    Sana-sprint: One-step diffusion with continuous-time consistency distillation,

    J. Chen, S. Xue, Y . Zhao, J. Yu, S. Paul, J. Chen, H. Cai, E. Xie, and S. Han, “Sana-sprint: One-step diffusion with continuous-time consistency distillation,” arXiv:2503.09641, 2025

    2025

  11. [17]

    Consistency trajectory models: Learning probability flow ODE trajectory of diffusion,

    D. Kim, C.-H. Lai, W.-H. Liao, N. Murata, Y . Takida, T. Uesaka, Y . He, Y . Mitsufuji, and S. Ermon, “Consistency trajectory models: Learning probability flow ODE trajectory of diffusion,” in ICLR, 2023

    2023

  12. [18]

    Presto! distilling steps and layers for accelerating music generation

    Z. Novack, G. Zhu, J. Casebeer, J. McAuley, T. Berg-Kirkpatrick, and N. J. Bryan, “Presto! distilling steps and layers for accelerating music generation.” in ICLR, 2025

    2025

  13. [19]

    One-step diffusion models with f- divergence distribution matching,

    Y . Xu, W. Nie, and A. Vahdat, “One-step diffusion models with f- divergence distribution matching,” arXiv:2502.15681, 2025

    2025 arXiv

  14. [20]

    Flow straight and fast: Learning to generate and transfer data with rectified flow,

    X. Liu, C. Gong, and Q. Liu, “Flow straight and fast: Learning to generate and transfer data with rectified flow,” arXiv:2209.03003, 2022

    2022 arXiv

  15. [21]

    Progressive distillation for fast sampling of diffusion models,

    T. Salimans and J. Ho, “Progressive distillation for fast sampling of diffusion models,” arXiv:2202.00512, 2022

    2022 arXiv

  16. [22]

    Distilling diffusion models into conditional gans,

    M. Kang, R. Zhang, C. Barnes, S. Paris, S. Kwak, J. Park, E. Shechtman, J.-Y . Zhu, and T. Park, “Distilling diffusion models into conditional gans,” arXiv:2405.05967, 2024

    2024 arXiv

  17. [23]

    CFG++: Manifold-constrained classifier free guidance for diffusion models,

    H. Chung, J. Kim, G. Y . Park, H. Nam, and J. C. Ye, “CFG++: Manifold-constrained classifier free guidance for diffusion models,” arXiv:2406.08070, 2024

    2024 arXiv

  18. [24]

    Ufogen: You forward once large scale text-to-image generation via diffusion gans,

    Y . Xu, Y . Zhao, Z. Xiao, and T. Hou, “Ufogen: You forward once large scale text-to-image generation via diffusion gans,” in CVPR, 2024

    2024

  19. [25]

    Diffusion ad- versarial post-training for one-step video generation,

    S. Lin, X. Xia, Y . Ren, C. Yang, X. Xiao, and L. Jiang, “Diffusion ad- versarial post-training for one-step video generation,” arXiv:2501.08316, 2025

    2025

  20. [27]

    Soundctm: Uniting score-based and consistency models for text-to-sound generation,

    K. Saito, D. Kim, T. Shibuya, C.-H. Lai, Z.-W. Zhong, Y . Takida, and Y . Mitsufuji, “Soundctm: Uniting score-based and consistency models for text-to-sound generation,” arXiv:2405.18503, 2024

    2024 arXiv

  21. [28]

    Improving musical accompaniment co-creation via diffusion transformers,

    J. Nistal, M. Pasini, and S. Lattner, “Improving musical accompaniment co-creation via diffusion transformers,” arXiv:2410.23005, 2024

    2024 arXiv

  22. [30]

    The gan is dead; long live the gan! a modern baseline gan,

    N. Huang, A. Gokaslan, V . Kuleshov, and J. Tompkin, “The gan is dead; long live the gan! a modern baseline gan,” in ICML Workshop on Structured Probabilistic Inference and Generative Modeling , 2024

    2024

  23. [31]

    The relativistic discriminator: a key element missing from standard gan,

    A. Jolicoeur-Martineau, “The relativistic discriminator: a key element missing from standard gan,” arXiv:1807.00734, 2018

    2018 arXiv

  24. [32]

    One step diffusion via shortcut models,

    K. Frans, D. Hafner, S. Levine, and P. Abbeel, “One step diffusion via shortcut models,” arXiv:2410.12557, 2024

    2024 arXiv

  25. [33]

    Large-scale contrastive language-audio pretraining with feature fusion and keyword-to-caption augmentation,

    Y . Wu, K. Chen, T. Zhang, Y . Hui, T. Berg-Kirkpatrick, and S. Dubnov, “Large-scale contrastive language-audio pretraining with feature fusion and keyword-to-caption augmentation,” in ICASSP, 2023

    2023

  26. [34]

    Noise-contrastive estimation: A new estimation principle for unnormalized statistical models,

    M. Gutmann and A. Hyv ¨arinen, “Noise-contrastive estimation: A new estimation principle for unnormalized statistical models,” in AISTATS, 2010

    2010

  27. [35]

    Fine-tuning language models from human preferences,

    D. M. Ziegler, N. Stiennon, J. Wu, T. B. Brown, A. Radford, D. Amodei, P. Christiano, and G. Irving, “Fine-tuning language models from human preferences,” arXiv:1909.08593, 2019

    1909 arXiv

  28. [36]

    Direct preference optimization: Your language model is secretly a reward model,

    R. Rafailov, A. Sharma, E. Mitchell, C. D. Manning, S. Ermon, and C. Finn, “Direct preference optimization: Your language model is secretly a reward model,” in NeurIPS, 2023

    2023

  29. [37]

    Deepseekmath: Pushing the limits of mathematical reasoning in open language models,

    Z. Shao, P. Wang, Q. Zhu, R. Xu, J. Song, X. Bi, H. Zhang, M. Zhang, Y . Li, Y . Wuet al., “Deepseekmath: Pushing the limits of mathematical reasoning in open language models,” arXiv:2402.03300, 2024

    2024 arXiv

  30. [38]

    Rank analysis of incomplete block designs: I. the method of paired comparisons,

    R. A. Bradley and M. E. Terry, “Rank analysis of incomplete block designs: I. the method of paired comparisons,” Biometrika, vol. 39, no. 3/4, pp. 324–345, 1952

    1952

  31. [39]

    Exploring the limits of transfer learning with a unified text-to-text transformer,

    C. Raffel, N. Shazeer, A. Roberts, K. Lee, S. Narang, M. Matena, Y . Zhou, W. Li, and P. J. Liu, “Exploring the limits of transfer learning with a unified text-to-text transformer,” Journal of machine learning research , vol. 21, no. 140, pp. 1–67, 2020

    2020

  32. [40]

    Query-key normalization for transformers,

    A. Henry, P. R. Dachapally, S. Pawar, and Y . Chen, “Query-key normalization for transformers,” arXiv:2010.04245, 2020

    2010 arXiv

  33. [41]

    Group normalization,

    Y . Wu and K. He, “Group normalization,” in ECCV, 2018

    2018

  34. [42]

    Sigmoid-weighted linear units for neural network function approximation in reinforcement learning,

    S. Elfwing, E. Uchibe, and K. Doya, “Sigmoid-weighted linear units for neural network function approximation in reinforcement learning,” Neural networks, vol. 107, pp. 3–11, 2018

    2018

  35. [43]

    Look, listen, and learn more: Design choices for deep audio embeddings,

    A. L. Cramer, H.-H. Wu, J. Salamon, and J. P. Bello, “Look, listen, and learn more: Design choices for deep audio embeddings,” in ICASSP, 2019

    2019

  36. [44]

    Efficient training of audio transformers with patchout,

    K. Koutini, J. Schl ¨uter, H. Eghbal-Zadeh, and G. Widmer, “Efficient training of audio transformers with patchout,” arXiv:2110.05069, 2021

    2021 arXiv

  37. [45]

    Distilling diversity and control in diffusion models,

    R. Gandikota and D. Bau, “Distilling diversity and control in diffusion models,” arXiv:2503.10637, 2025

    2025

  38. [46]

    Diff- a-riff: Musical accompaniment co-creation via latent diffusion models,

    J. Nistal, M. Pasini, C. Aouameur, M. Grachten, and S. Lattner, “Diff- a-riff: Musical accompaniment co-creation via latent diffusion models,” arXiv:2406.08384, 2024

    2024 arXiv

  39. [47]

    Audiocaps: Generating captions for audios in the wild,

    C. D. Kim, B. Kim, H. Lee, and G. Kim, “Audiocaps: Generating captions for audios in the wild,” in NAACL, 2019

    2019

  40. [48]

    webmushra—a comprehensive framework for web-based listening tests,

    M. Schoeffler, S. Bartoschek, F.-R. St¨oter, M. Roess, S. Westphal, B. Edler, and J. Herre, “webmushra—a comprehensive framework for web-based listening tests,” Journal of Open Research Software , vol. 6, no. 1, 2018

    2018

  41. [49]

    Adversarial permutation invariant training for universal sound separation,

    E. Postolache, J. Pons, S. Pascual, and J. Serr `a, “Adversarial permutation invariant training for universal sound separation,” in ICASSP, 2023

    2023

  42. [50]

    On loss functions and evaluation metrics for music source separation,

    E. Gus ´o, J. Pons, S. Pascual, and J. Serr `a, “On loss functions and evaluation metrics for music source separation,” in ICASSP, 2022

    2022

  43. [51]

    Consistency models made easy,

    Z. Geng, A. Pokle, W. Luo, J. Lin, and J. Z. Kolter, “Consistency models made easy,” arXiv:2406.14548, 2024

    2024 arXiv

Pith tools

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