Pith. sign in

REVIEW 3 major objections 3 minor 31 references

Conditional Latent Diffusion-Based Speech Enhancement Via Dual Context Learning

T0 review · 3 major / 3 minor · reviewed 2026-08-10 · deepseek-v4-flash

Pith's one-line read Conditional latent diffusion with dual-context learning outperforms prior diffusion-based speech enhancement methods, achieving PESQ 2.71 at one-fifth the inference cost.

desk verdict A plausible speed-up for diffusion speech enhancement, but the DCL ablation is confounded by training data volume, so the paper's causal claim about dual-context learning is not yet supported. read the letter →

arxiv 2501.10052 v1 pith:666OESPR submitted 2025-01-17 cs.SD eess.AS

classification cs.SDeess.AS
keywords speechenhancementlatentdiffusionmodeldual-contextlearningmel-spectrogramVAEunseennoisegeneralizationtext-conditionedgenerative
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 speech enhancement can be made cheaper and more robust by moving diffusion-based generation into a low-dimensional latent space and by training the model to generate not only clean speech but also the background noise it removes. Existing diffusion enhancers operate on full waveforms or complex spectrograms, making each reverse step expensive, and most of them model only the clean-speech target. The authors compress noisy, clean, and noise-only mel-spectrograms with a shared VAE, run a conditional diffusion model on the resulting 8-channel latent, and use text-prompted dual tasks so the same network can generate either the speech latent or the noise latent. They report that the result outperforms the compared diffusion-based methods on seen and unseen noise, does so at roughly one-fifth the real-time factor, and generalizes to out-of-domain test sets.

What carries the argument

The central mechanism is dual-context learning (DCL): one shared conditional latent diffusion model is trained on two tasks selected at random during training, either generating the clean-speech latent or generating the background-noise latent, with the task selected by a text embedding such as 'Speech enhancement' versus 'Background noise estimation.' A variational autoencoder compresses 64-band mel-spectrograms into an 8-channel latent with time and frequency downsampled by $r=4$, and a frozen neural vocoder maps the decoded mel-spectrogram back to the waveform. A U-Net with cross-attention to the text embedding estimates the added noise, conditioned on the noisy latent by channel-wise concatenation, and this shared speech-and-noise modelling is what carries the generalization and efficiency claims.

What would settle it

Encode a clean mel-spectrogram with the VAE encoder, decode it with the VAE decoder, and render it with the vocoder, then compare PESQ and ESTOI of that reconstruction with the cLDM+DCL enhanced output of the same utterance. If the reconstruction already matches or exceeds the reported 2.71 PESQ, the diffusion module is not the source of the quality gain; if it is much worse, the gap quantifies how much headroom the latent codec leaves.

Watch

Extended reading notes

Core claim

The paper claims that running conditional denoising diffusion in a compressed latent space, rather than on raw waveforms or spectrograms, preserves or improves enhancement quality while cutting inference cost, and that training the same diffusion model to generate both clean-speech latents and background-noise latents, distinguished only by a text instruction, makes the enhancer more discriminative and more robust to unseen noises. On its seen-noise test set the model reaches PESQ 2.71, ESTOI 0.88, SI-SDR 17.1 dB, WV-MOS 3.73, and DNS-MOS 3.54 with a real-time factor of 0.40 at 50 reverse steps, beating three earlier diffusion-based methods on every reported metric at about one-fifth their compute. On all five held-out noise types and on two mismatched public benchmark test sets, the model's PESQ remains close to its seen-noise score, which the authors attribute to the noise distribution learned through dual-context learning.

Load-bearing premise

The compressed audio representation used as the diffusion target must lose nothing that matters for speech quality, and the paper never reports how faithfully the compressor alone reproduces clean speech.

Editorial extensions

If this is right

  • At 50 reverse steps the model's real-time factor of 0.40 is about one-fifth that of the score-based spectrogram diffusion baselines, so latent-domain diffusion can run in near real time while improving PESQ and the other reported metrics.
  • At just 10 reverse steps the dual-context model matches the 50-step model without dual-context learning, indicating that learning the noise distribution reduces the number of denoising iterations needed for a given quality.
  • DCL outperforms conditioning on features from a pre-trained noise classifier, so modelling the full noise distribution rather than noise-type labels is claimed to be the more effective way to exploit noise information.
  • On unseen noise types and mismatched public test sets the model's PESQ stays close to its seen-noise score, whereas the discriminative baselines drop noticeably, supporting the claim of robustness to unmatched acoustic conditions.

Reading between the lines

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

  • Not tested in the paper, but directly implied by the design, is that the frozen VAE plus vocoder sets a quality ceiling: a reconstruction benchmark that encodes a clean mel-spectrogram, decodes it, and renders it would separate the diffusion model's contribution from the codec's contribution.
  • Because the two generation contexts are chosen by text embeddings, one could prompt the same model at inference time with instructions such as 'remove babble' or 'keep the music' and obtain different enhancement behavior without retraining; only the two fixed prompts are evaluated.
  • The reported generalization may be partly a data-scale effect, since training used 250 hours of noise and thousands of hours of noisy pairs; on smaller datasets the DCL margin over noise-classifier conditioning might shrink.
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 / 3 minor

Summary. The paper proposes a speech enhancement system, cLDM+DCL, that combines a conditional latent diffusion model (cLDM) with dual-context learning (DCL). A frozen VAE compresses mel-spectrograms into a low-dimensional latent space, and a frozen BigVGAN vocoder reconstructs waveforms from the enhanced mel-spectrogram. The diffusion model is trained to reverse the noising process for both clean-speech latents and background-noise latents, conditioned on the noisy latent and a T5 text embedding that selects the generation target ("Speech enhancement" or "Background noise estimation"). At inference only the speech-enhancement prompt is used. Experiments compare cLDM+DCL against CDiffuSE, SGMSE+, StoRM, cLDM+NASE, and two discriminative baselines on seen noise, unseen noise, VoiceBank+DEMAND, and DNS Challenge 2020 test sets, reporting PESQ, ESTOI, SI-SDR, WV-MOS, DNS-MOS, and RTF. The authors report that cLDM+DCL outperforms the compared diffusion methods on all metrics at 50 reverse steps while running at a substantially lower RTF, and that DCL improves generalization to unseen noise types.

Significance. If the claims hold, the paper makes a useful practical contribution: it demonstrates that latent-domain diffusion with a compact VAE representation can match or improve upon waveform/spectrogram-domain diffusion methods for speech enhancement while reducing the real-time factor by a factor of roughly 5 compared to SGMSE+ and StoRM. The dual-context learning idea, sharing one model across speech and noise generation with text prompts, is well motivated and consistent with current latent-diffusion practice. The experimental design has strengths: diffusion baselines are retrained on the same training set, the step-count ablation is reported, and several out-of-domain benchmarks are included. However, the central attribution of the gain to DCL is not yet supported by a clean controlled comparison, and the reconstruction ceiling of the frozen VAE/vocoder is unquantified, so the significance of the method relative to non-latent diffusion baselines remains somewhat uncertain.

major comments (3)
  1. [§3.4, §3.5, Table 2] The only controlled evidence for DCL is the ablation in Table 2, but that comparison changes two variables simultaneously: cLDM+DCL is trained on 3,000 h of noisy-clean pairs plus 1,000 h of noisy-noise pairs (Section 3.4), while the cLDM baseline is trained solely on 3,000 h of noisy-clean pairs (Section 3.5). The observed gains at 50 steps (PESQ 2.64 to 2.71, ESTOI 0.86 to 0.88, SI-SDR 16.4 to 17.1) could therefore come from the extra training data and the additional noise-estimation task rather than from dual-context conditioning per se. Moreover, Table 4 contains no cLDM-only row, so the claim that DCL improves generalization to unseen noise is unsupported by a data-matched comparison. Please add a controlled ablation: either train cLDM on the same 4,000 h of data without the dual-context objective, or train cLDM+DCL on 3,000 h only, and report unseen-noise results for both configurations.
  2. [§2.4, §3.4, Table 1] The manuscript never quantifies the reconstruction ceiling imposed by the frozen VAE and vocoder. The VAE compresses a 64-band mel-spectrogram into an 8-channel latent with time and frequency downsampled by 4 (Table 1), discarding phase before the vocoder, yet no mel/VAE reconstruction metrics or oracle experiments are reported. Without such a ceiling, the comparison against waveform/spectrogram diffusion baselines (Tables 3, 5, 6) cannot separate the benefit of latent-space diffusion from the smoothing/reconstruction properties of the vocoder. Please report the reconstruction quality of clean speech through the VAE decoder and BigVGAN, e.g., PESQ/ESTOI/SI-SDR of the vocoder output versus the original clean signal, and if possible an oracle experiment where the clean latent z_X is decoded.
  3. [§3.2, Tables 3–6] All headline differences among cLDM+DCL, cLDM+NASE, and the diffusion baselines are small in absolute terms (e.g., PESQ 2.71 vs 2.66 on seen noise, 2.95 vs 2.91 on DNS), and no error bars, confidence intervals, or significance tests are reported. In addition, the unseen-noise comparison in Table 4 reports only PESQ, so the "superior generalization capability" claim rests on a single metric. Please report variance or significance tests across test items and full-metric results for unseen noise, or temper the generalization claim accordingly.
minor comments (3)
  1. [§2.2, Eqs. (2)–(3)] Equation (2) writes the variance as (1 − ᾱt)ϵ where ϵ is the noise sample; it should be (1 − ᾱt)I. Equation (3) writes the coefficient of ϵ as (1 − ᾱt); it should be sqrt(1 − ᾱt). Please correct these typos because they affect the closed-form sampling formula central to the method.
  2. [§3.4, Table 2] The cLDM-only row in Table 2 is reported only at T=50, so the claim that cLDM without DCL at 50 steps performs similarly to cLDM+DCL at 10 steps is a single-point comparison. Reporting the same step sweep for cLDM-only would make the interaction between DCL and step count clearer.
  3. [Throughout] Please fix typographical issues such as "V AE", "V ocoder", "V oiceBank+DEMAND", and "Traning Details" in Section 3.4; these are cosmetic but distract from the presentation.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: cLDM+DCL is an empirical systems paper whose central claims are evaluated against external baselines; the DCL ablation, though potentially confounded by training-data volume, is not a derivation that reduces to its own inputs.

full rationale

The paper's derivation chain consists of standard conditional diffusion equations (Eqs. 1-7, adopted from Ho et al. and Rombach et al.), a frozen VAE/vocoder pipeline, and a DCL training scheme. The central claims are empirical: cLDM+DCL is compared against CDiffuSE, SGMSE+, StoRM, cLDM+NASE, Conv-TasNet, and MetricGAN+ on seen, unseen, VoiceBank+DEMAND, and DNS Challenge 2020 test sets. None of the reported metrics are defined in terms of the model parameters, and no fitted quantity is renamed as a prediction. The only self-citation, Ref. [13] (Audit, which shares an author), is used as inspiration for applying latent diffusion to audio, not as load-bearing justification for the enhancement results. The strongest possible concern is the DCL ablation in Table 2: Section 3.1 states 'the noisy-noise pairs are only used for the dual-context learning framework,' so cLDM+DCL sees 4,000h of training data while the cLDM ablation sees only 3,000h. This is a real experimental confound that weakens the causal claim that DCL itself drives the gain, but it is not circularity: the gain is measured, not derived, and the comparison does not reduce by construction to an input. No uniqueness theorem is imported from the authors, no ansatz is smuggled via self-citation, and no known result is merely renamed. The paper's contributions are self-contained empirical findings with an internal-validity caveat, so the appropriate circularity score is 0.

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

The method rests on standard diffusion math plus several domain assumptions: mel-spectrograms and a compressed VAE latent are sufficient representations, simulated noisy pairs transfer to real conditions, and the T5 prompt pair cleanly separates the two generation tasks. The main hand-chosen design parameters are the compression ratio, the number of reverse steps, the prompt texts, and the training SNR range.

free parameters (4)
  • VAE compression ratio r = 4
    Chosen by hand; defines latent space dimensions (Table 1) and trades generation speed against reconstruction fidelity, central to the latent approach.
  • Number of reverse diffusion steps T = 50
    Chosen for final evaluation (Section 3.4); ablation in Table 2 shows quality improves with T, so the chosen operating point affects reported metrics and RTF.
  • DCL text prompts = "Speech enhancement" and "Background noise estimation"
    Hand-chosen text conditions (Section 2.3) that distinguish the two generation tasks; the entire DCL scheme depends on them.
  • Training SNR range = -5 to 15 dB
    Uniformly sampled SNR for noisy-clean and noisy-noise pair generation (Section 3.1); determines the difficulty distribution of the training data.
assumptions (5)
  • standard math Standard DDPM forward/reverse process as in Ho et al. 2020 (Eqs. 1-6)
    The conditional diffusion formulation and the reweighted loss in Eq. (7) are taken from prior work without re-derivation.
  • domain assumption Mel-spectrogram domain retains sufficient information for speech enhancement
    All processing operates on 64-band mel-spectrograms; phase is discarded and reconstructed by BigVGAN (Section 2.4), so any information lost in the mel representation cannot be recovered.
  • domain assumption VAE latent space is a valid low-dimensional proxy for speech and noise content
    The VAE (Section 2.4) compresses mel-spectrograms by factor 4 in time and frequency; the paper assumes this latent retains enough detail for the diffusion model to enhance speech, without reporting reconstruction metrics.
  • domain assumption Additive noise model y = x + n and simulated training pairs generalize to real-world conditions
    Training data is generated by mixing LibriSpeech with filtered AudioSet noise at random SNRs (Section 3.1); the OOD test sets (VoiceBank+DEMAND, DNS2020) rely on this transfer.
  • domain assumption T5 text embeddings provide a task-distinguishing conditioning signal for the shared denoiser
    Cross-attention to T5 embeddings (Section 2.3) is assumed to route the same U-Net to either speech or noise generation; no analysis of prompt separation is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Conditional Latent Diffusion-Based Speech Enhancement Via Dual Context Learning." pith.science (2026). https://pith.science/paper/666OESPR

@misc{pith2026250110052,
  author       = {Pith},
  title        = {Pith review of: Conditional Latent Diffusion-Based Speech Enhancement Via Dual Context Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/666OESPR}},
  note         = {Machine review of arXiv:2501.10052}
}
read the original abstract

Recently, the application of diffusion probabilistic models has advanced speech enhancement through generative approaches. However, existing diffusion-based methods have focused on the generation process in high-dimensional waveform or spectral domains, leading to increased generation complexity and slower inference speeds. Additionally, these methods have primarily modelled clean speech distributions, with limited exploration of noise distributions, thereby constraining the discriminative capability of diffusion models for speech enhancement. To address these issues, we propose a novel approach that integrates a conditional latent diffusion model (cLDM) with dual-context learning (DCL). Our method utilizes a variational autoencoder (VAE) to compress mel-spectrograms into a low-dimensional latent space. We then apply cLDM to transform the latent representations of both clean speech and background noise into Gaussian noise by the DCL process, and a parameterized model is trained to reverse this process, conditioned on noisy latent representations and text embeddings. By operating in a lower-dimensional space, the latent representations reduce the complexity of the generation process, while the DCL process enhances the model's ability to handle diverse and unseen noise environments. Our experiments demonstrate the strong performance of the proposed approach compared to existing diffusion-based methods, even with fewer iterative steps, and highlight the superior generalization capability of our models to out-of-domain noise datasets (https://github.com/modelscope/ClearerVoice-Studio).

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 20 canonical work pages

  1. [1]

    Conditional Latent Diffusion-Based Speech Enhancement Via Dual Context Learning

    INTRODUCTION Speech enhancement, referring to the task of restoring clean speech from noise-corrupted speech signals, has been seen in many appli- cations in telecommunication and robust speech recognition. Re- cent advancements in speech enhancement have been achieved by leveraging deep learning techniques. There are two typical types of speech enhanceme...

  2. [2]

    Speech enhancement

    METHOD 2.1. System Overview The process of speech enhancement involves estimating clean speech x from noisy speech y := x + n, where n is the background noise. The overall architecture of our proposed system is illustrated in Fig. 1. The system primarily consists of a variational auto- encoder (V AE), a conditional latent diffusion model (cLDM), and a voc...

  3. [3]

    EXPERIMENT 3.1. Dataset Since generative models benefit from large-scale training data, we selected the LibriSpeech corpus [17], using the ’train-clean-360’ subset for training clean speech and ’test-clean’ for testing speech. The training set comprises 360 hours of speech. For noise data, we utilized the AudioSet corpus [18] and filtered out audio clips ...

  4. [4]

    The proposed cLDM operates in a low- dimensional latent space, reducing complexity and improving the efficiency of the generation process

    CONCLUSIONS We introduced a generative framework, cLDM+DCL, by integrat- ing a conditional latent diffusion model with dual-context learning for speech enhancement. The proposed cLDM operates in a low- dimensional latent space, reducing complexity and improving the efficiency of the generation process. The DCL scheme further strengthens the model’s abilit...

  5. [5]

    Conv-tasnet: Surpassing ideal time–frequency magnitude masking for speech separation,

    Y . Luo and N. Mesgarani, “Conv-tasnet: Surpassing ideal time–frequency magnitude masking for speech separation,” IEEE/ACM transactions on audio, speech, and language pro- cessing, vol. 27, no. 8, pp. 1256–1266, 2019

  6. [6]

    MetricGAN+: An improved version of Metric- GAN for speech enhancement,

    S.-W. Fu, C. Yu, T.-A. Hsieh, P. Plantinga, M. Ravanelli, X. Lu, and Y . Tsao, “MetricGAN+: An improved version of Metric- GAN for speech enhancement,” in Proc. of Interspeech, 2022

  7. [7]

    Conditional diffusion probabilistic model for speech enhancement,

    Y .-J. Lu, Z.-Q. Wang, S. Watanabe, A. Richard, C. Yu, and Y . Tsao, “Conditional diffusion probabilistic model for speech enhancement,” in Proc. of ICASSP, 2022

  8. [8]

    Speech enhancement and dereverberation with diffusion-based generative models,

    J. Richter, S. Welker, J.-M. Lemercier, B. Lay, and T. Gerk- mann, “Speech enhancement and dereverberation with diffusion-based generative models,” IEEE Trans. Audio, Speech, Language Proc., 2023

Show all 31 references
  1. [9]

    StoRM: A diffusion-based stochastic regeneration model for speech enhancement and dereverberation,

    J.-M. Lemercier, J. Richter, S. Welker, and T. Gerkmann, “StoRM: A diffusion-based stochastic regeneration model for speech enhancement and dereverberation,” arXiv:2212.11851, 2022

  2. [10]

    Deep unsupervised learning using nonequilibrium ther- modynamics,

    J. Sohl-Dickstein, E.Weiss, N. Maheswaranathan, and S. Gan- guli, “Deep unsupervised learning using nonequilibrium ther- modynamics,” in Proc. of ICML, 2015

  3. [11]

    Denoising diffusion probabilis- tic models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilis- tic models,” arXiv:2006.11239, 2020

  4. [12]

    Improved denoising diffusion probabilistic models,

    A. Nichol and P. Dhariwal, “Improved denoising diffusion probabilistic models,” arXiv:2102.09672, 2021

  5. [13]

    Nadiffuse: Noise-aware diffusion-based model for speech enhancement,

    W. Wang, D. Yang, Q. Ye, B. Cao, and Y . Zou, “Nadiffuse: Noise-aware diffusion-based model for speech enhancement,” arXiv:2309.01212, 2023

  6. [14]

    Noise-aware speech enhancement using diffusion probabilistic model,

    Y . Hu, C. Chen, R. Li, Q. Zhu, and E. S. Chng, “Noise-aware speech enhancement using diffusion probabilistic model,” arXiv:2307.08029, 2024

  7. [15]

    High-resolution image synthesis with latent diffusion models,

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Om- mer, “High-resolution image synthesis with latent diffusion models,” arXiv:2112.10752, 2022

  8. [16]

    Audioldm: Text-to-audio generation with latent diffusion models,

    H. Liu, Z. Chen, Y . Yuan, X. Mei, X. Liu, D. Mandic, W. Wang, and M. D. Plumbley, “Audioldm: Text-to-audio generation with latent diffusion models,” arXiv:2301.12503, 2023

  9. [17]

    Audit: Audio editing by following instructions with latent dif- fusion models,

    Y . Wang, Z. Ju, X. Tan, L. He, Z. Wu, J. Bian, and S. Zhao, “Audit: Audio editing by following instructions with latent dif- fusion models,” arXiv:2304.00830, 2023

  10. [18]

    Exploring the lim- its 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 lim- its of transfer learning with a unified text-to-text transformer,” Journal of Machine Learning Research, 2020

  11. [19]

    Auto-encoding variational bayes,

    D. P. Kingma and M. Welling, “Auto-encoding variational bayes,” arXiv:1312.6114, 2022

  12. [20]

    Bigvgan: A universal neural vocoder with large-scale train- ing,

    S. Lee, W. Ping, B. Ginsburg, B. Catanzaro, and S. Yoon, “Bigvgan: A universal neural vocoder with large-scale train- ing,” arXiv:2206.04658, 2023

  13. [21]

    Lib- rispeech: an ASR corpus based on public domain audio books,

    V . Panayotov, G. Chen, D. Povey, and S. Khudanpur, “Lib- rispeech: an ASR corpus based on public domain audio books,” in Proc. of ICASSP, 2015, pp. 5206–5210

  14. [22]

    Audio Set: an ontology and human-labeled dataset for audio events,

    J.F. Gemmeke, D.P.W. Ellis, D. Freedman, A. Jansen, W. Lawrence, R.C. Moore, M. Plakal, and M. Ritter, “Audio Set: an ontology and human-labeled dataset for audio events,” in Proc. of ICASSP, 2017, pp. 776–780

  15. [23]

    Investigating RNN-based speech enhancement methods for noise-robust Text-to-Speech,

    C.Valentini-Botinhao, X.Wang, S.Takaki, and J.Yamagishi, “Investigating RNN-based speech enhancement methods for noise-robust Text-to-Speech,” in Proc. of SSW , 2016, p. 146–152

  16. [24]

    The interspeech 2020 deep noise suppres- sion challenge: Datasets, subjective testing framework, and challenge results,

    C. K. Reddy et al., “The interspeech 2020 deep noise suppres- sion challenge: Datasets, subjective testing framework, and challenge results,” arXiv preprint arXiv:2005.13981, 2020

  17. [25]

    Perceptual evaluation of speech quality (PESQ) - a new method for speech quality assessment of telephone networks and codecs,

    A. Rix, J. Beerends, M. Hollier, and A. Hekstra, “Perceptual evaluation of speech quality (PESQ) - a new method for speech quality assessment of telephone networks and codecs,” inProc. of ICASSP, 2001

  18. [26]

    An algorithm for predicting the in- telligibility of speech masked by modulated noise maskers,

    J. Jensen and C. H. Taal, “An algorithm for predicting the in- telligibility of speech masked by modulated noise maskers,” IEEE Trans. on Audio, Speech, and Language Proc. (TASLP) , vol. 24, no. 11, pp. 2009–2022, 2016

  19. [27]

    SDR – half-baked or well done?,

    J. L. Roux, S. Wisdom, H. Erdogan, and J. R. Hershey, “SDR – half-baked or well done?,” in Proc. of ICASSP, 2019

  20. [28]

    HIFI++: A unified framework for bandwidth extension and speech en- hancement,

    P. Andreev, A. Alanov, O. Ivanov, and D. Vetrov, “HIFI++: A unified framework for bandwidth extension and speech en- hancement,” in Proc. of ICASSP, 2023

  21. [29]

    DNSMOS p.835: A non-intrusive perceptual objective speech quality metric to evaluate noise suppressors,

    K.A.R. Chandan, G. Vishak, and C. Ross, “DNSMOS p.835: A non-intrusive perceptual objective speech quality metric to evaluate noise suppressors,” arXiv:2110.01763, 2022

  22. [30]

    Beats: Audio pre-training with acoustic tokenizers,

    S. Chen, Y . Wu, C. Wang, S. Liu, D. Tompkins, Z. Chen, and F. Wei, “Beats: Audio pre-training with acoustic tokenizers,” arXiv:2212.09058, 2022

  23. [31]

    Decoupled weight decay regularization,

    Ilya Loshchilov and Frank Hutter, “Decoupled weight decay regularization,” arXiv: 1711.05101, 2019

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.