Pith. sign in

REVIEW 4 major objections 5 minor 42 references

SpeechRefiner: Towards Perceptual Quality Refinement for Front-End Algorithms

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A single post-processing model trained with conditional flow matching can repair the artifacts left by a wide range of speech front-ends, including systems it never saw during training.

desk verdict A useful, honestly scoped task-agnostic speech refiner whose central perceptual claim rests on a single objective metric; worth reviewing, but it needs a subjective or intelligibility check. read the letter →

arxiv 2506.13709 v1 pith:LN2T6B4S submitted 2025-06-16 eess.AS cs.SD

Sirui Li , Shuai Wang , Zhijun Liu , Zhongjie Jiang , Yannan Wang , Haizhou Li This is my paper · ORCID
classification eess.AScs.SD
keywords speechrestorationperceptualqualityconditionalflowmatchingfront-endprocessingenhancementdereverberationseparationtargetspeakerextraction
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 sets out to show that a single post-processing model can repair the residual noise and artifacts that speech front-ends leave behind, even when those front-ends were never seen during training. SpeechRefiner is a conditional flow matching model that takes the distorted output of any front-end and regenerates a cleaner mel-spectrogram, then resynthesizes the waveform with a neural vocoder. The authors benchmark it against task-specific refiners on enhancement, dereverberation, and separation, and also test it on an internal pipeline combining several front-ends. They report consistent SIGMOS improvements across all tasks, including on audio-only and audio-visual target speaker extraction systems that were absent from training. The motivation is that objective metrics like SI-SNR miss perceptual artifacts, and a general-purpose refiner could improve real-world pipelines at scale.

What carries the argument

The engine is optimal-transport conditional flow matching (OT-CFM), a variant of flow matching in which a neural network learns the conditional vector field $u_t(x_0|x_1)=x_1-(1-\sigma_{\min})x_0$ along the linear interpolation $\phi_t(x)=(1-(1-\sigma_{\min})t)x_0+tx_1$, with $x_1$ a clean speech mel-spectrogram and $c$ the distorted speech used as a conditioning signal. The vector field is parameterized by a 10-block Conformer with rotary position embeddings, followed by Conv2D and ResBlock2D layers, and the refined mel-spectrogram is synthesized by the Vocos neural vocoder. This setup lets the model regenerate speech from the distribution of clean speech given arbitrary front-end outputs, without needing front-end-specific inputs or joint training.

What would settle it

Run a controlled listening test (for example MUSHRA or the ITU-T P.804 subjective procedure that SIGMOS approximates) on the same test utterances from Tables 2 and 3, comparing unprocessed front-end outputs with SpeechRefiner outputs. If human raters do not show a significant preference for SpeechRefiner, or if the ordering of systems by SIGMOS diverges from the ordering by human ratings, the paper's central claim is not supported.

Watch

Extended reading notes

Core claim

SpeechRefiner is a standalone audio restoration module trained with optimal-transport conditional flow matching (OT-CFM): given a distorted speech signal c, the model learns a time-dependent vector field that transports Gaussian noise to the distribution of clean mel-spectrograms, conditioned on c, and a pre-trained Vocos vocoder turns the predicted mel-spectrograms back into waveforms. Trained primarily on speech corrupted by a single simulated impairment source, the model is then applied to outputs of CDiffuSE, SGMSE+, Sepformer, Spex+, and MuSE, as well as an internal denoising-AEC-dereverberation pipeline. The reported results show that SpeechRefiner matches or exceeds task-specific refinement systems such as Diffiner+, StoRM, and Fast-GeCo, and that it improves SIGMOS scores on all evaluated front-ends, including unseen target speaker extraction systems. The paper interprets this as evidence that front-end artifacts share common structure that a single generative refiner can learn to remove.

Load-bearing premise

SIGMOS scores are treated as a faithful stand-in for human perceptual quality, and every claim of perceptual improvement in the paper is measured by this non-intrusive metric rather than by listening tests.

Editorial extensions

If this is right

  • A single universal refiner can replace task-specific refinement modules, so a new front-end algorithm can be paired with SpeechRefiner without retraining or co-design.
  • SpeechRefiner can be inserted at the end of automated speech-data cleaning pipelines to salvage audio that quality filters would otherwise discard.
  • Because the model improves SIGMOS on both enhancement and dereverberation outputs when trained on a different internal pipeline, front-end artifact correction appears to transfer across impairment types.
  • The baseline comparisons show that task-specific refiners can degrade sharply on mismatched tasks, while SpeechRefiner's performance remains consistent.

Reading between the lines

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

  • If SIGMOS is replaced by a subjective listening test, the reported margins may shrink; a direct MUSHRA or ITU-T P.804 listening study on the same outputs would settle whether the improvements are perceptually real.
  • The same flow-matching design could be extended to other artifact classes not tested here, such as codec compression, background music, or text-to-speech synthesis artifacts, since the model only requires paired distorted-clean speech.
  • The paper's reported limitation that multi-speaker inputs get merged into one voice suggests a natural extension: conditioning on speaker embeddings or attempting speaker-wise flow trajectories, which the paper does not explore.
  • In practical deployment, the 64-step Euler solver may be a latency bottleneck; distillation or fewer-step ODE solvers could make the refiner real-time, something the paper does not address.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes SpeechRefiner, a post-processing module that uses optimal-transport conditional flow matching with a Conformer-based mel-spectrogram predictor and the Vocos neural vocoder to refine the outputs of various speech front-end algorithms (enhancement, dereverberation, separation, and target speaker extraction). The authors evaluate SpeechRefiner against task-specific refinement baselines (Diffiner+, StoRM, Fast-GeCo) and on an internal multi-stage front-end pipeline, using the SIGMOS non-intrusive metric as the sole evaluation criterion. The central claim is that a single task-agnostic post-processor significantly improves perceptual quality and generalizes to unseen front-end algorithms. The paper also reports a limitation in Section 5: on severely distorted, low-intelligibility inputs, the model tends to produce clean but semantically meaningless audio.

Significance. If the central claim is substantiated, a single task-agnostic refinement module that improves perceived quality across diverse front-ends would be practically valuable, particularly for automated data-cleaning pipelines. The paper includes several strengths: the OT-CFM formulation is clearly described, the comparison includes recent task-specific baselines, the use of a standardized objective metric (SIGMOS) allows reproducibility, and the authors provide audio demos. The generalization experiment across unseen front-ends (including Spex+ and MuSE) is a useful stress test. However, the evidence base is too narrow to establish the perceptual-quality claim: all conclusions rest on SIGMOS point estimates, and the acknowledged failure mode of meaningless-but-clean outputs is exactly the kind of error that SIGMOS may not penalize. The claim that SpeechRefiner consistently outperforms or matches task-specific baselines is also contradicted by the dereverberation results in Table 2.

major comments (4)
  1. [Section 3.4; Tables 2-3; Section 5] The central claim of significant perceptual quality improvement rests entirely on SIGMOS scores, without any subjective listening test or intelligibility metric. Section 5 explicitly acknowledges that on severely distorted inputs, SpeechRefiner 'tends to produce clean yet semantically meaningless audio signals.' SIGMOS is a non-intrusive metric with no explicit semantic or intelligibility component, so such outputs can plausibly receive high SIG and OVRL scores. The evaluation must demonstrate that the metric tracks human perception on these exact outputs, for example by adding an intelligibility measure (e.g., ASR word error rate or STOI) or a listening test on a subset, especially in the severe-distortion regime the authors themselves identify.
  2. [Table 2, Dereverberation row] The paper claims SpeechRefiner 'outperforms or matches existing methods in most metrics,' but the dereverberation results show SpeechRefiner underperforming StoRM on SIG (3.60 vs 3.69) and OVRL (3.20 vs 3.23). This is a direct counterexample to the claim of consistent improvement over task-specific refiners. The authors should either revise the claim to acknowledge this exception or analyze why SpeechRefiner regresses on this task, since the central message of consistent advantage is otherwise weakened.
  3. [Section 3.2; Table 1] The training data for SpeechRefiner is never explicitly specified. The statement that all baseline models were trained and evaluated under 'identical data conditions' as SpeechRefiner is ambiguous: it is not clear whether one multi-task SpeechRefiner model was trained on a mixture of front-end outputs, or separate per-task models were trained and evaluated independently. The composition and size of the training set, as well as the relationship between the model used in Table 2 and the internal-dataset model used in Section 4.2, must be stated for the generalization claims to be interpretable and reproducible.
  4. [Tables 2-3] All reported scores are point estimates over 500 utterances per test set, with no confidence intervals, error bars, or significance tests. Several differences are small (e.g., OVRL 3.20 vs 3.23 for dereverberation), so claims of 'significant improvements' and 'outperforms or matches' are not statistically supported. The authors should report variance or perform significance testing, at least for the headline comparisons.
minor comments (5)
  1. [Section 2.2.1] The word 'non-trival' should be 'non-trivial'.
  2. [Figure 1 caption] The caption refers to 'SpeechRefiner Augmented Speech Restoration System,' which may be clearer if harmonized with the model name 'SpeechRefiner' used throughout the text.
  3. [Section 4.1] The sentence 'its performance was significantly poor, with an overall score of 2.20' should specify that this is the OVRL score, matching the metrics reported in Table 2.
  4. [Table 3] The row labeled 'Enhancement & Dereverb: Internal' does not describe which specific front-end algorithms compose the internal pipeline; a footnote referencing Section 4.2 would make the table self-contained.
  5. [Section 3.3] The number of ODE inference steps is fixed at 64; the authors could report sensitivity to this hyperparameter, as it directly affects the inference cost and possibly the quality of the refined output.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the evaluation is empirical, externally benchmarked against SIGMOS and independent baselines; the reported gains are not inputs by construction.

full rationale

SpeechRefiner's reported contribution is an empirical training-and-evaluation pipeline, not a closed-form derivation. The model is trained with the OT-CFM loss in Eq. (3), mapping distorted speech c to clean mel-spectrograms x1, and evaluated with SIGMOS (Section 3.4), an externally published objective metric that is not defined in terms of SpeechRefiner or its loss. The central generalization claim (Section 4.2, Table 3) is tested on front-end outputs not used in training, including Spex+ and MuSE; although these two front-ends originate from the same research group, they are independently published processors used as input generators, not as evidence for SpeechRefiner's success, so the self-citation is not load-bearing. All baseline systems are also processed through the same Vocos vocoder to ensure consistency, so the comparison is not rigged by the vocoder choice. The acknowledged limitation in Section 5 (clean yet semantically meaningless outputs on severely distorted input) is a validity risk for the SIGMOS-only evaluation, but it is a correctness concern, not circularity: the paper nowhere fits a parameter to SIGMOS or defines the metric in terms of the model. No fitted input is renamed as a prediction, and no uniqueness theorem or ansatz is imported from the authors' prior work. The derivation chain is therefore self-contained with respect to circularity.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The model's central claim rests primarily on the SIGMOS evaluation metric and on the sufficiency of mel-spectrogram representations, both of which are domain assumptions rather than fitted parameters. The few hand-chosen hyperparameters (sigma_min, ODE steps, conformer depth) affect performance but are not themselves predictors.

free parameters (3)
  • sigma_min = 0
    Flow matching hyperparameter set to zero in experiments (Section 2.2.1); simplifies the OT path and affects the learned mapping.
  • ODE inference steps = 64
    Number of Euler steps used at inference (Section 3.3), following prior work; affects output quality versus compute.
  • number of conformer blocks = 10
    Architecture depth chosen by the authors (Section 2.2.2); no ablation is provided to justify the choice.
assumptions (4)
  • standard math Conditional flow matching provides a valid generative training objective for mel-spectrogram prediction.
    Assumed from Lipman et al. (2021) [21]; the paper relies on OT-CFM theory without re-deriving it.
  • domain assumption Mel-spectrogram representation is sufficient for high-quality speech restoration, given a neural vocoder.
    The model only predicts mels and uses Vocos; any information lost in the mel representation cannot be recovered (Section 2.3).
  • domain assumption SIGMOS is a reliable proxy for human perceptual quality.
    All quality claims are based on SIGMOS (Section 3.4); no listening test is conducted.
  • domain assumption Training on the internal front-end pipeline (denoising, AEC, dereverberation) generalizes to other front-ends such as separation and TSE.
    The key generalization result in Table 3 relies on this transfer assumption (Section 4.2).

how reviews work

0 comments
Cite this review

Pith. "Pith review of SpeechRefiner: Towards Perceptual Quality Refinement for Front-End Algorithms." pith.science (2026). https://pith.science/paper/LN2T6B4S

@misc{pith2026250613709,
  author       = {Pith},
  title        = {Pith review of: SpeechRefiner: Towards Perceptual Quality Refinement for Front-End Algorithms},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LN2T6B4S}},
  note         = {Machine review of arXiv:2506.13709}
}
read the original abstract

Speech pre-processing techniques such as denoising, de-reverberation, and separation, are commonly employed as front-ends for various downstream speech processing tasks. However, these methods can sometimes be inadequate, resulting in residual noise or the introduction of new artifacts. Such deficiencies are typically not captured by metrics like SI-SNR but are noticeable to human listeners. To address this, we introduce SpeechRefiner, a post-processing tool that utilizes Conditional Flow Matching (CFM) to improve the perceptual quality of speech. In this study, we benchmark SpeechRefiner against recent task-specific refinement methods and evaluate its performance within our internal processing pipeline, which integrates multiple front-end algorithms. Experiments show that SpeechRefiner exhibits strong generalization across diverse impairment sources, significantly enhancing speech perceptual quality. Audio demos can be found at https://speechrefiner.github.io/SpeechRefiner/.

Figures

Figures reproduced from arXiv: 2506.13709 by the authors.

Figure 1
Figure 1. Overview of the SpeechRefiner Augmented Speech Restoration System. SpeechRefiner acts as a post-processing module to address the distortions and artifacts introduced by front-end processing algorithms [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The architecture of the conformer-based model. 2.2. Conformer-based Model 2.2.1. Optimal-Transport Conditional Flow Matching Modeling data distributions in high-dimensional spaces is non￾trival for neural networks, especially when the target true distri￾bution q(x) is unknown. One feasible solution is transforming a simple prior distribution p0, such as a Gaussian, into a more complex target distribution p1, which a… view at source ↗
Figure 3
Figure 3. Spectrogram Visualization of SpeechRefiner a) and b) show the outputs of the SGMSE+ and AV-TSE systems, re￾spectively; c) and d) depict the corresponding restored versions produced by SpeechRefiner. 5. Conclusion In this paper, we proposed SpeechRefiner, a post-processing module aimed at mitigating impairments and artifacts intro￾duced by various front-end processing algorithms, which signif￾icantly affect perceptua… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 34 canonical work pages

  1. [1]

    Introduction Speech pre-processing front-ends are critical for intelligent sys- tems functioning in real-world acoustic environments, where target human voices are often corrupted by reverberation, noise, and overlapping sound sources. Typical tasks involve enhance- ment techniques to reduce non-speech interference, such as de-noising [1] and de-reverbera...

    work page

  2. [2]

    SpeechRefiner In this work, we propose SpeechRefiner, a standalone refine- ment module designed to address degradation caused by various speech processing front-ends. 2.1. Model overview SpeechRefiner consists of two main components, as shown in Figure 1. First, a Conformer-based model predicts mel- spectrograms from the distorted speech output of the fro...

    work page Pith review arXiv 2025

  3. [3]

    Front-end processing systems We aim to test our system on speech processed by different front-end algorithms to evaluate its generalization performance

    Experimental Setups 3.1. Front-end processing systems We aim to test our system on speech processed by different front-end algorithms to evaluate its generalization performance. For each task, we randomly select 500 utterances from the cor- responding test set to evaluate the proposed SpeechRefiner. The related front-end algorithms and the corresponding t...

    work page

  4. [4]

    Comparison with front-end-specific baselines We first evaluate SpeechRefiner against state-of-the-art refin- ers designed for specific front-end tasks

    Results and Analysis 4.1. Comparison with front-end-specific baselines We first evaluate SpeechRefiner against state-of-the-art refin- ers designed for specific front-end tasks. Table 2 compares the performance across three core tasks: speech enhancement, dereverberation, and separation. Our SpeechRefiner outper- forms or matches existing methods in most ...

    work page

  5. [5]

    Trained on data derived from a single impairment source, SpeechRefiner demonstrated re- markable generalization capabilities across a variety of unseen front-end algorithms

    Conclusion In this paper, we proposed SpeechRefiner, a post-processing module aimed at mitigating impairments and artifacts intro- duced by various front-end processing algorithms, which signif- icantly affect perceptual quality. Trained on data derived from a single impairment source, SpeechRefiner demonstrated re- markable generalization capabilities ac...

    work page

  6. [6]

    62401377, Shenzhen Science and Technology Program (Shenzhen Key Laboratory, Grant No

    Acknowledgements This work was supported by National Natural Science Foun- dation of China, Grant No. 62401377, Shenzhen Science and Technology Program (Shenzhen Key Laboratory, Grant No. ZDSYS20230626091302006), Shenzhen Science and Technol- ogy Research Fund (Fundamental Research Key Project, Grant No. JCYJ20220818103001002), Program for Guangdong In- t...

    work page

  7. [7]

    Tasnet: time-domain audio separa- tion network for real-time, single-channel speech separation,

    Y . Luo and N. Mesgarani, “Tasnet: time-domain audio separa- tion network for real-time, single-channel speech separation,” in ICASSP 2018. IEEE, 2018, pp. 696–700

    work page 2018

  8. [8]

    RaD-Net: A Repairing and Denoising Network for Speech Signal Improvement

    M. Liu, Z. Chen, X. Yan, Y . Lv, X. Xia, C. Huang, Y . Xiao, and L. Xie, “Rad-net: A repairing and denoising network for speech signal improvement,” arXiv preprint arXiv:2401.04389, 2024

    work page Pith review arXiv 2024

Show all 42 references
  1. [9]

    A review of multi-objective deep learning speech denoising methods,

    A. Azarang and N. Kehtarnavaz, “A review of multi-objective deep learning speech denoising methods,” Speech Communica- tion, vol. 122, pp. 1–10, 2020

    2020

  2. [10]

    P. A. Naylor and N. D. Gaubitch, Speech dereverberation . Springer Science & Business Media, 2010

    2010

  3. [11]

    Supervised speech separation based on deep learning: An overview,

    D. Wang and J. Chen, “Supervised speech separation based on deep learning: An overview,” IEEE/ACM transactions on audio, speech, and language processing, vol. 26, no. 10, pp. 1702–1726, 2018

    2018

  4. [12]

    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 processing, vol. 27, no. 8, pp. 1256–1266, 2019

    2019

  5. [13]

    Neural target speech extraction: An overview,

    K. Zmolikova, M. Delcroix, T. Ochiai, K. Kinoshita, J. ˇCernock`y, and D. Yu, “Neural target speech extraction: An overview,”IEEE Signal Processing Magazine, vol. 40, no. 3, pp. 8–29, 2023

    2023

  6. [14]

    Sdr– half-baked or well done?

    J. Le Roux, S. Wisdom, H. Erdogan, and J. R. Hershey, “Sdr– half-baked or well done?” in ICASSP 2019. IEEE, 2019, pp. 626–630

    2019

  7. [15]

    Miipher: A ro- bust speech restoration model integrating self-supervised speech and text representations,

    Y . Koizumi, H. Zen, S. Karita, Y . Ding, K. Yatabe, N. Morioka, Y . Zhang, W. Han, A. Bapna, and M. Bacchiani, “Miipher: A ro- bust speech restoration model integrating self-supervised speech and text representations,” in2023 IEEE Workshop on Applications of Signal Processing...

    2023

  8. [16]

    The AdamW optimizer is used with a learning rate of 2 × 10−4, β1 = 0.9 and β2 = 0.99. A cosine annealing scheduler 2https://github.com/neillu23/CDiffuSE 3https://github.com/sp-uhh/sgmse 4https://github.com/WangHelin1997/Fast-GeCo 5https://github.com/haoheliu/voicefixer is appl...

    2024

  9. [17]

    Rad-net 2: A causal two-stage repairing and denois- ing speech enhancement network with knowledge distillation and complex axial self-attention,

    ——, “Rad-net 2: A causal two-stage repairing and denois- ing speech enhancement network with knowledge distillation and complex axial self-attention,” in Proc. Interspeech 2024 , 2024, pp. 1700–1704

    2024

  10. [18]

    Autoprep: An automatic preprocessing framework for in-the-wild speech data,

    J. Yu, H. Chen, Y . Bian, X. Li, Y . Luo, J. Tian, M. Liu, J. Jiang, and S. Wang, “Autoprep: An automatic preprocessing framework for in-the-wild speech data,” in ICASSP 2024. IEEE, 2024, pp. 1136–1140

    2024

  11. [19]

    Wenetspeech4tts: A 12,800-hour mandarin tts corpus for large speech generation model bench- mark,

    L. Ma, D. Guo, K. Song, Y . Jiang, S. Wang, L. Xue, W. Xu, H. Zhao, B. Zhang, and L. Xie, “Wenetspeech4tts: A 12,800-hour mandarin tts corpus for large speech generation model bench- mark,” in Proc. Interspeech 2024, 2024, pp. 1840–1844

    2024

  12. [20]

    Dnsmos: A non-intrusive perceptual objective speech quality metric to evaluate noise sup- pressors,

    C. K. Reddy, V . Gopal, and R. Cutler, “Dnsmos: A non-intrusive perceptual objective speech quality metric to evaluate noise sup- pressors,” in ICASSP 2021. IEEE, 2021, pp. 6493–6497

    2021

  13. [21]

    The loss function for CFM is expressed as: LCF M(θ) =Et,q(x1),pt(x|x1)∥vt(x; θ) − ut(x|x1)∥2

    offers a solution by replacing the original vector field ut with a conditional vector field ut(x|x1), making the problem more manageable. The loss function for CFM is expressed as: LCF M(θ) =Et,q(x1),pt(x|x1)∥vt(x; θ) − ut(x|x1)∥2. (2) Here, pt(x|x1) represents the conditional...

  14. [22]

    Icassp 2024 speech signal improvement challenge,

    N.-C. Ristea, A. Saabas, R. Cutler, B. Naderi, S. Braun, and S. Branets, “Icassp 2024 speech signal improvement challenge,” in 2024 IEEE International Conference on Acoustics, Speech, and Signal Processing Workshops (ICASSPW). IEEE, 2024, pp. 15– 16

    2024

  15. [23]

    V oicefixer: A unified framework for high-fidelity speech restoration,

    H. Liu, X. Liu, Q. Kong, Q. Tian, Y . Zhao, D. Wang, C. Huang, and Y . Wang, “V oicefixer: A unified framework for high-fidelity speech restoration,” in Proc. Interspeech 2022, 2022, pp. 4232– 4236

    2022

  16. [24]

    Storm: A diffusion-based stochastic regeneration model for speech en- hancement and dereverberation,

    J.-M. Lemercier, J. Richter, S. Welker, and T. Gerkmann, “Storm: A diffusion-based stochastic regeneration model for speech en- hancement and dereverberation,”IEEE/ACM Transactions on Au- dio, Speech, and Language Processing, 2023

    2023

  17. [25]

    Diffiner: A versatile diffusion-based generative refiner for speech enhancement,

    R. Sawata, N. Murata, Y . Takida, T. Uesaka, T. Shibuya, S. Taka- hashi, and Y . Mitsufuji, “Diffiner: A versatile diffusion-based generative refiner for speech enhancement,” in Proc. Interspeech 2023, 2023, pp. 3824–3828

    2023

  18. [26]

    Diffusion-based signal refiner for speech separation,

    M. Hirano, K. Shimada, Y . Koyama, S. Takahashi, and Y . Mitsu- fuji, “Diffusion-based signal refiner for speech separation,” arXiv preprint arXiv:2305.05857, 2023

    2023

  19. [27]

    Separate and diffuse: Us- ing a pretrained diffusion model for better source separation,

    S. Lutati, E. Nachmani, and L. Wolf, “Separate and diffuse: Us- ing a pretrained diffusion model for better source separation,” in 12th International Conference on Learning Representations, ICLR 2024, 2024

    2024

  20. [28]

    Noise-robust speech separation with fast generative correction,

    H. Wang, J. Villalba, L. Moro-Velazquez, J. Hai, T. Thebaud, and N. Dehak, “Noise-robust speech separation with fast generative correction,” in Proc. Interspeech 2024, 2024, pp. 2165–2169

    2024

  21. [29]

    Flow matching for generative modeling,

    Y . Lipman, R. T. Chen, H. Ben-Hamu, M. Nickel, and M. Le, “Flow matching for generative modeling,” in 11th International Conference on Learning Representations, ICLR 2023, 2023

    2023

  22. [30]

    Conformer: Convolution- augmented transformer for speech recognition,

    A. Gulati, J. Qin, C.-C. Chiu, N. Parmar, Y . Zhang, J. Yu, W. Han, S. Wang, Z. Zhang, Y . Wu et al. , “Conformer: Convolution- augmented transformer for speech recognition,” in Proc. Inter- speech 2020, 2020, pp. 5036–5040

    2020

  23. [31]

    Roformer: Enhanced transformer with rotary position embedding,

    J. Su, M. Ahmed, Y . Lu, S. Pan, W. Bo, and Y . Liu, “Roformer: Enhanced transformer with rotary position embedding,” Neuro- computing, vol. 568, p. 127063, 2024

    2024

  24. [32]

    Train short, test long: At- tention with linear biases enables input length extrapolation,

    O. Press, N. A. Smith, and M. Lewis, “Train short, test long: At- tention with linear biases enables input length extrapolation,” in ICLR, 2022

    2022

  25. [33]

    V ocos: Closing the gap between time-domain and fourier-based neural vocoders for high-quality audio synthesis,

    H. Siuzdak, “V ocos: Closing the gap between time-domain and fourier-based neural vocoders for high-quality audio synthesis,” arXiv preprint arXiv:2306.00814, 2023

    2023 arXiv

  26. [34]

    Conditional diffusion probabilistic model for speech en- hancement,

    Y .-J. Lu, Z.-Q. Wang, S. Watanabe, A. Richard, C. Yu, and Y . Tsao, “Conditional diffusion probabilistic model for speech en- hancement,” in ICASSP 2022. IEEE, 2022, pp. 7402–7406

    2022

  27. [35]

    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 SSW, 2016, pp. 146–152

    2016

  28. [36]

    Speech enhancement and dereverberation with diffusion-based generative models,

    J. Richter, S. Welker, J.-M. Lemercier, B. Lay, and T. Gerkmann, “Speech enhancement and dereverberation with diffusion-based generative models,” IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 31, pp. 2351–2364, 2023

    2023

  29. [37]

    Attention is all you need in speech separation,

    C. Subakan, M. Ravanelli, S. Cornell, M. Bronzi, and J. Zhong, “Attention is all you need in speech separation,” inICASSP 2021. IEEE, 2021, pp. 21–25

    2021

  30. [38]

    Librimix: An open-source dataset for generalizable speech separation,

    J. Cosentino, M. Pariente, S. Cornell, A. Deleforge, and E. Vin- cent, “Librimix: An open-source dataset for generalizable speech separation,” arXiv preprint arXiv:2005.11262, 2020

    2005 arXiv

  31. [39]

    Spex+: A complete time domain speaker extraction network,

    M. Ge, H. Li, C. Xu, L. Wang, E. S. Chng, and J. Dang, “Spex+: A complete time domain speaker extraction network,”Interspeech 2020, 2020

    2020

  32. [40]

    Muse: Multi-modal target speaker extraction with visual cues,

    Z. Pan, R. Tao, C. Xu, and H. Li, “Muse: Multi-modal target speaker extraction with visual cues,” in ICASSP 2021. IEEE, 2021, pp. 6678–6682

    2021

  33. [41]

    Libritts-r: A restored multi-speaker text-to-speech corpus,

    Y . Koizumi, H. Zen, S. Karita, Y . Ding, K. Yatabe, N. Morioka, M. Bacchiani, Y . Zhang, W. Han, and A. Bapna, “Libritts-r: A restored multi-speaker text-to-speech corpus,” arXiv preprint arXiv:2305.18802, 2023

    2023 arXiv

  34. [42]

    Grad-tts: A diffusion probabilistic model for text-to-speech,

    V . Popov, I. V ovk, V . Gogoryan, T. Sadekova, and M. Kudinov, “Grad-tts: A diffusion probabilistic model for text-to-speech,” in International Conference on Machine Learning . PMLR, 2021, pp. 8599–8608

    2021

Pith tools

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