Pith. sign in

REVIEW 5 major objections 4 minor 1 cited by

Improving Resource-Efficient Speech Enhancement via Neural Differentiable DSP Vocoder Refinement

T0 review · 5 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

Pith's one-line read A compact feature-prediction network feeding a zero-phase DDSP vocoder improves speech enhancement quality by up to 19% (DNSMOS) and intelligibility by up to 4% (STOI) with little added compute.

desk verdict A plausible and useful DDSP-refinement method for efficient SE, with real non-causal gains, but the abstract's compute claim is unmeasured and the causal results contradict the paper's own text. read the letter →

arxiv 2508.14709 v1 pith:NKFCYXNX submitted 2025-08-20 eess.AS cs.SD

classification eess.AScs.SD
keywords speechenhancementDDSPvocodersource-filtersynthesislow-resourcemodelsreal-timeacousticfeaturepredictionDNSMOSSTOI
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 proposes that a speech-enhancement system for wearable devices can get a large perceptual-quality gain by adding a differentiable DSP vocoder as the final resynthesis stage, while keeping the neural part small. Instead of having a network output a waveform directly, a compact model predicts three clean acoustic parameters from noisy speech—fundamental frequency, periodicity, and spectral envelope—and a zero-phase DDSP vocoder synthesizes the enhanced utterance. Trained end-to-end with STFT and adversarial losses, the added vocoder improves DNSMOS by up to 19% and STOI by up to 4% over the same backbones without it, at an extra cost of about 16K parameters and no significant increase in computation. This matters because it points to a way of obtaining natural-sounding, intelligible speech on resource-limited devices without the computational burden of neural vocoders.

What carries the argument

The load-bearing mechanism is the zero-phase DDSP vocoder, a source-filter resynthesizer with no trainable parameters. It decomposes each frame into a periodic signal—an impulse train at the predicted fundamental frequency, weighted by periodicity and the spectral envelope—and an aperiodic noise signal, then sums them in the frequency domain: s = iFFT(P × V) * eimp(F0) + iFFT((1 − P) × V × Enoise). Because the filter is zero-phase, the vocoder avoids explicit phase modeling and stays computationally cheap, while end-to-end differentiability lets the feature-prediction network be trained directly on waveform-level objectives.

What would settle it

Train the same backbones with clean features extracted by a different off-the-shelf F0/envelope estimator, or with oracle features produced by the vocoder's own analysis; if the DNSMOS/STOI gains shrink or reverse, the reported improvement is tied to the particular reference-feature extractor rather than to general speech enhancement.

Watch

Extended reading notes

Core claim

The central claim is that a two-stage 'predict-then-resynthesize' design beats direct predictive enhancement on low-resource models. The paper's system consists of a lightweight backbone (a convolutional transformer network or a convolutional recurrent network, at roughly 300K or 600K parameters) that maps noisy, reverberant speech to a spectral envelope, a 12-band periodicity vector, and a fundamental-frequency track. These features are fed to a zero-phase DDSP vocoder, which rebuilds the waveform through a source-filter model: an impulse train and white noise are weighted by periodicity, shaped by the envelope, and combined in the frequency domain. Since both the feature predictor and voco

Load-bearing premise

The approach assumes that clean acoustic features (F0, periodicity, spectral envelope) can be reliably extracted from clean speech and used as training targets for a small predictor, but the paper does not specify how those reference features are obtained.

Editorial extensions

If this is right

  • On the non-causal CTN with magnitude input, adding the DDSP vocoder raises DNSMOS overall from 2.72 to 3.23 (small model) and from 2.83 to 3.37 (large model), and STOI from 73.0 to 75.3 and 75.1 to 78.3—so the quality gains are not tied to one model size.
  • Mel-cepstral distortion drops sharply with the vocoder (e.g., 6.85 to 5.89 for the small CTN), indicating the synthesized output is closer to the reference speech's acoustic structure.
  • A causal CRN with 8 ms algorithmic latency also improves with the vocoder (small variant DNSMOS from 2.79 to 3.15), showing the approach is compatible with real-time streaming constraints.
  • Scaling the CTN from about 300K to 2.5M parameters produces larger relative gains when the vocoder is present, so the resynthesis path makes better use of extra capacity.
  • Complex-STFT input does not help DDSP-based models; the zero-phase vocoder already determines the phase, so magnitude input is the better design choice.

Reading between the lines

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

  • The paper's gains hinge on the reference features used as training targets. If those features come from a clean-speech analyzer that systematically differs from the DDSP vocoder's own internal analysis, end-to-end training may partly learn to compensate for that mismatch rather than to suppress noise; comparing against an alternative off-the-shelf feature extractor would settle this.
  • The predict-and-resynthesize structure is not limited to enhancement: the same feature-prediction backbone could be redirected to bandwidth extension, voice conversion, or other speech-synthesis front ends, since the interface to the vocoder is just three acoustic parameters.
  • DNSMOS, a non-intrusive metric, is the main evidence for the quality gains; a formal listening test on device-captured noise would test whether the reported 19% improvement translates to human preference.
  • The near-zero parameter cost of DDSP suggests a possible next step: shrinking the feature predictor further or distilling a larger enhancement model into the compact predictor, since the vocoder already supplies much of the generative structure.
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

5 major / 4 minor

Summary. The paper proposes a speech-enhancement framework that combines a compact neural feature predictor with a fixed, differentiable DDSP vocoder. The predictor estimates a spectral envelope, fundamental frequency, and periodicity from noisy speech; the vocoder synthesizes the enhanced waveform from these features. Training uses MSE on F0 and periodicity, a multi-resolution STFT loss, and an adversarial loss. The authors compare CTN and CRN backbones with and without the DDSP refinement on a DNS2020 subset, reporting PESQ, STOI, DNSMOS, MCD, and SSIM. The abstract claims a 4% STOI and 19% DNSMOS improvement over strong baselines without significant added computation, suitable for real-time use.

Significance. If the claims hold, the work would be a useful contribution to resource-constrained speech enhancement: it shows that a parameter-free, differentiable vocoder can be attached to a small predictor to improve perceived quality, and the use of multiple backbones and model scales gives some evidence of generality. Strengths include the end-to-end differentiability of the pipeline, the use of external (non-circular) metrics such as PESQ/STOI/DNSMOS, and the explicit reporting of parameter counts. However, the resource-efficiency and real-time claims are not substantiated by any compute or latency measurements, the causal CRN results are inconsistent with the abstract's blanket statement, and the training target extraction is under-specified. These issues are load-bearing and prevent acceptance in their current form.

major comments (5)
  1. [Abstract; Section III-B] The claim 'without significantly increasing computation' is not supported by any measurement in the paper. No FLOPs, MACs, latency, or real-time factor is reported; parameter count alone (roughly 10–20K increase) is not a proxy for compute. The CRN+DDSP variant adds a 4-layer causal Transformer (d_model=64, FF=96), and the DDSP vocoder itself performs STFT, iFFT, and source-filter operations per frame, none of which are quantified. The 24× reduction mentioned in Section I is relative to the NHV vocoder, not the total system. Please add system-level computational measurements or substantially qualify the claim.
  2. [Section IV-B; Table II] The statement that the causal CRN results show 'consistent improvements across both intrusive and non-intrusive metrics' is contradicted by Table II: for the larger CRN, PESQ is 1.85→1.81 and STOI is 72.75→72.36, both worse than the baseline. The abstract's headline 4% STOI gain is taken from the non-causal CTN in Table I (75.10→78.25), not from the causal real-time setting. The text should separate the non-causal and causal stories, report the negative results transparently, and adjust any blanket claims about intelligibility/quality improvements.
  3. [Section II-F; Section II-B] The training targets for F0 and periodicity are said to be 'reference extracted from the clean signal,' but the extraction method is never specified. This is load-bearing: if the features are extracted with the DDSP vocoder's own analysis, the predictor may learn to match the synthesizer's internal representation rather than independent speech features; if a standard algorithm is used, the choice affects reproducibility and the validity of the MSE losses. The hyperparameters α and β in Eq. (2) are also not given. Please specify the exact feature extractor(s) and hyperparameter values, or otherwise clarify the source of the references.
  4. [Section III-B] The comparison is not controlled for optimization budget. Models with DDSP are trained for 250 epochs with a warmup/decay schedule and an adversarial loss, while the non-DDSP baselines are trained for 100 epochs with a constant learning rate and, apparently, no adversarial loss. The text says 100 epochs 'was found to be sufficient for convergence,' but no ablation is shown. If the baselines are under-trained or trained with a different objective, the with/without-DDSP comparison conflates the vocoder contribution with longer training and the presence of a discriminator. Please either match the training budgets/protocols or provide evidence that the baseline is at its optimum.
  5. [Section IV-A] The central perceptual-quality claim rests on DNSMOS, since the authors themselves argue that intrusive metrics like PESQ/STOI may be unreliable for generative models. While this is a reasonable caveat, the paper does not include a listening test, and DNSMOS can be sensitive to synthetic timbre. The DNSMOS gains could partly reflect the vocoder's characteristic sound rather than better noise suppression. A small subjective test, or at least an analysis showing DNSMOS gains correspond to reduced noise (e.g., via BAK improvements or spectrogram inspection), would substantiate the quality claim.
minor comments (4)
  1. [Eq. (1)] The notation P × V is ambiguous: P is a p-dimensional mel-bandwise vector, V is v-dimensional, and the dimensions are not aligned before multiplication. Please clarify the broadcasting/linear mapping used to combine these features.
  2. [Abstract/IV-C] The percentages '4% (STOI) and 19% (DNSMOS)' should be labeled as relative or absolute improvements. The STOI gain in Table I is relative (75.10→78.25), and this is not stated. Also, Figure 3 has no numeric values in the text or caption; for a scaling study, a table would be more informative.
  3. [Section II-D] The description of the locally-connected CNN says 'each frequency bin of the mel-spectrogram is treated as an individual channel,' but the channel count is then given as 4. This is confusing; please clarify the input shaping and channel dimensions.
  4. [Section II-C/II-D] The paper does not state the STFT window size, hop length, and FFT size used for the input representation, nor the mel-band count before the CNN. These parameters are needed for reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found; external metrics and standard losses make the derivation self-contained. The unmeasured compute claim and unspecified feature extractor are evidence gaps, not circular reductions.

full rationale

The derivation chain is not circular. The system is a supervised feature-prediction network followed by a fixed differentiable DDSP vocoder, trained with MSE on F0/periodicity, multi-resolution STFT loss, and an adversarial loss (Eq. 2-4). None of these losses is identical to the reported evaluation metrics (PESQ, STOI, DNSMOS, MCD, SSIM), and the metrics are computed on the final waveform against clean references and standard baselines (Tables I-II). The vocoder is adopted from prior work (DDSP [15], [18]) rather than re-derived here; the 24x computation-reduction comparison to NHV is imported from [18], which is externally checkable prior work and is not used to compute any of the reported metric gains. The only mild self-referential element is that the clean acoustic features used as training targets are extracted from clean speech using an unspecified method; even if that extractor were the vocoder's own analysis, this would be an autoencoder-style representation rather than a circular prediction of the evaluation metrics. The paper's unmeasured 'without significantly increasing computation' claim and the missing feature-extractor specification are substantive reproducibility/evidence gaps, but they are not circular reductions: no equation reduces to its own input. Therefore no circular step is identified.

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

The method is an empirical system whose conclusions rest on the reliability of the DDSP synthesis module and on the chosen training and evaluation objectives. The main free parameters are the un-reported loss weights and the manually chosen STFT loss settings.

free parameters (5)
  • alpha (loss weight for F0 MSE)
    Weight for the L2 loss on predicted F0 in Eq. (2); reported as alpha but never given a value.
  • beta (loss weight for periodicity MSE)
    Weight for the L2 loss on predicted periodicity in Eq. (2); reported as beta but never given a value.
  • Multi-resolution STFT window sizes W = [512, 1024, 2048]
    Window sizes in samples chosen by the authors for the multi-resolution STFT loss.
  • Multi-resolution STFT loss weights Lambda = [25.7, 51.3, 102.5]
    Weights for each resolution in Eq. (3); chosen by hand.
  • Number of discriminators K = 16
    From the original DDSP vocoder setup, not optimized in this paper.
assumptions (5)
  • domain assumption Source-filter model of speech production: speech is generated from periodic (voiced) and aperiodic (unvoiced) excitations shaped by a vocal tract filter.
    The whole DDSP vocoder synthesis (Eq. 1) rests on this speech production model, cited to classical literature.
  • domain assumption The DDSP vocoder is a differentiable, zero-phase synthesis module that produces high-quality speech with no trainable parameters.
    The paper relies entirely on the quality and differentiability of the vocoder from ref [18]; no verification is done in this paper.
  • domain assumption Clean acoustic features (F0, periodicity, spectral envelope) can be accurately extracted and used as training targets.
    Training uses MSE between predicted features and references extracted from clean signals (Section II-F). The extraction method is not specified.
  • domain assumption The multi-resolution STFT and adversarial losses are suitable objectives for training a generative speech enhancement model.
    These losses are adopted from prior work without justification specific to this setting.
  • domain assumption DNSMOS, PESQ, STOI, MCD, and SSIM are valid proxies for perceptual quality and intelligibility.
    The conclusions are based on these metrics; no human listening test is performed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Improving Resource-Efficient Speech Enhancement via Neural Differentiable DSP Vocoder Refinement." pith.science (2026). https://pith.science/paper/NKFCYXNX

@misc{pith2026250814709,
  author       = {Pith},
  title        = {Pith review of: Improving Resource-Efficient Speech Enhancement via Neural Differentiable DSP Vocoder Refinement},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NKFCYXNX}},
  note         = {Machine review of arXiv:2508.14709}
}
read the original abstract

Deploying speech enhancement (SE) systems in wearable devices, such as smart glasses, is challenging due to the limited computational resources on the device. Although deep learning methods have achieved high-quality results, their computational cost limits their feasibility on embedded platforms. This work presents an efficient end-to-end SE framework that leverages a Differentiable Digital Signal Processing (DDSP) vocoder for high-quality speech synthesis. First, a compact neural network predicts enhanced acoustic features from noisy speech: spectral envelope, fundamental frequency (F0), and periodicity. These features are fed into the DDSP vocoder to synthesize the enhanced waveform. The system is trained end-to-end with STFT and adversarial losses, enabling direct optimization at the feature and waveform levels. Experimental results show that our method improves intelligibility and quality by 4% (STOI) and 19% (DNSMOS) over strong baselines without significantly increasing computation, making it well-suited for real-time applications.

Figures

Figures reproduced from arXiv: 2508.14709 by the authors.

Figure 1
Figure 1. Overview of the proposed system. The acoustic feature predictor estimates periodicity, spectral envelope, and fundamental frequency from noisy input, [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Simplified schema for the Source-Filter [23] model. [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. CTN (non-causal) performance across model scales. Results highlight [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

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. An Interpretable, Controllable Time-Varying IIR Denoiser for On-Device Assistive Hearing

    cs.SD 2026-03 reject novelty 6.0 of 10

    A neural controller that predicts time-varying biquad filter coefficients can denoise speech with an interpretable equalizer chain, though the abstract and body report inconsistent specifications.

Reference graph

Works this paper leans on

38 extracted references · 30 canonical work pages · cited by 1 Pith paper

  1. [1]

    P. C. Loizou, Speech enhancement: theory and practice . CRC press, 2007

  2. [2]

    A simple rnn model for lightweight, low-compute and low-latency multichannel speech enhancement in the time domain,

    A. Pandey, K. Tan, and B. Xu, “A simple rnn model for lightweight, low-compute and low-latency multichannel speech enhancement in the time domain,” in Proc. Interspeech 2023 , 2023, pp. 2478–2482

  3. [3]

    Decoupled spatial and temporal processing for resource efficient multichannel speech enhancement,

    A. Pandey and B. Xu, “Decoupled spatial and temporal processing for resource efficient multichannel speech enhancement,” in ICASSP 2024 - 2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2024, pp. 12 206–12 210

  4. [4]

    Ultra low-compute complex spectral masking for multichannel speech enhancement,

    A. Pandey and J. Azcarreta, “Ultra low-compute complex spectral masking for multichannel speech enhancement,” in ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2025, pp. 1–5

  5. [5]

    Dynamic gated recurrent neural network for compute-efficient speech enhancement,

    L. Cheng, A. Pandey, B. Xu, T. Delbruck, and S.-C. Liu, “Dynamic gated recurrent neural network for compute-efficient speech enhancement,” in Proc. INTERSPEECH 2024 , 2024

  6. [6]

    Modulating state space model with slowfast framework for compute- efficient ultra low-latency speech enhancement,

    L. Cheng, A. Pandey, B. Xu, T. Delbruck, V . K. Ithapu, and S.-C. Liu, “Modulating state space model with slowfast framework for compute- efficient ultra low-latency speech enhancement,” in ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2025, pp. 1–5

  7. [7]

    Low-latency speech enhancement via speech token generation,

    H. Xue, X. Peng, and Y . Lu, “Low-latency speech enhancement via speech token generation,” in ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2024, pp. 661–665

  8. [8]

    Genhancer: High-fidelity speech enhancement via generative modeling on discrete codec tokens,

    H. Yang, J. Su, M. Kim, and Z. Jin, “Genhancer: High-fidelity speech enhancement via generative modeling on discrete codec tokens,” in Interspeech 2024 , 2024, pp. 1170–1174

Show all 38 references
  1. [9]

    Ditse: High- fidelity generative speech enhancement via latent diffusion transform- ers,

    H. R. Guimar ˜aes, J. Su, R. Kumar, T. H. Falk, and Z. Jin, “Ditse: High- fidelity generative speech enhancement via latent diffusion transform- ers,” arXiv preprint arXiv:2504.09381 , 2025

  2. [10]

    Universal score-based speech enhancement with high content preservation,

    R. Scheibler, Y . Fujita, Y . Shirahata, and T. Komatsu, “Universal score-based speech enhancement with high content preservation,” in Interspeech 2024 , 2024, pp. 1165–1169

  3. [11]

    Flowse: Flow matching- based speech enhancement,

    S. Lee, S. Cheong, S. Han, and J. W. Shin, “Flowse: Flow matching- based speech enhancement,” in ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2025, pp. 1–5

  4. [12]

    Hifi-gan: Generative adversarial networks for efficient and high fidelity speech synthesis,

    J. Kong, J. Kim, and J. Bae, “Hifi-gan: Generative adversarial networks for efficient and high fidelity speech synthesis,” Advances in neural information processing systems , vol. 33, pp. 17 022–17 033, 2020

  5. [13]

    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,” in The Twelfth International Conference on Learning Representations , 2024. [Online]. Available: https://openreview.net/forum?id=vY9nzQmQBw

  6. [14]

    Spectrum-aware neural vocoder based on self-supervised learning for speech enhancement,

    Y . Song, D. Kim, H.-G. Kang, and N. Madhu, “Spectrum-aware neural vocoder based on self-supervised learning for speech enhancement,” in 2024 32nd European Signal Processing Conference (EUSIPCO) , 2024, pp. 16–20

  7. [15]

    Ddsp: Differentiable digital signal processing,

    J. Engel, L. H. Hantrakul, C. Gu, and A. Roberts, “Ddsp: Differentiable digital signal processing,” in International Conference on Learning Representations , 2020. [Online]. Available: https://openreview.net/forum?id=B1x1ma4tDr

  8. [16]

    O’shaughnessy, Speech communications: Human and machine (IEEE)

    D. O’shaughnessy, Speech communications: Human and machine (IEEE). Universities press, 1987

  9. [17]

    Speech enhancement—a review of modern meth- ods,

    D. O’Shaughnessy, “Speech enhancement—a review of modern meth- ods,” IEEE Transactions on Human-Machine Systems , vol. 54, no. 1, pp. 110–120, 2024

  10. [18]

    Ultra-lightweight neural differential dsp vocoder for high quality speech synthesis,

    P. Agrawal, T. Koehler, Z. Xiu, P. Serai, and Q. He, “Ultra-lightweight neural differential dsp vocoder for high quality speech synthesis,” in ICASSP 2024 - 2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , 2024, pp. 10 066–10 070

  11. [19]

    Neural homomorphic vocoder,

    Z. Liu, K. Chen, and K. Yu, “Neural homomorphic vocoder,” in Interspeech 2020 , 2020, pp. 240–244

  12. [20]

    Efficient speech enhancement with neural homomorphic synthesis,

    W. Jiang, T. Liu, and K. Yu, “Efficient speech enhancement with neural homomorphic synthesis,” in Interspeech 2022 , 2022, pp. 986–990

  13. [21]

    Speech enhancement with neural homomorphic synthesis,

    W. Jiang, Z. Liu, K. Yu, and F. Wen, “Speech enhancement with neural homomorphic synthesis,” in ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , 2022, pp. 376–380

  14. [22]

    Speech enhancement with integration of neural homomorphic synthesis and spectral masking,

    W. Jiang and K. Yu, “Speech enhancement with integration of neural homomorphic synthesis and spectral masking,” IEEE/ACM Transactions on Audio, Speech, and Language Processing , vol. 31, pp. 1758–1770, 2023

  15. [23]

    Fant, Acoustic theory of speech production: with calculations based on X-ray studies of Russian articulations

    G. Fant, Acoustic theory of speech production: with calculations based on X-ray studies of Russian articulations . Walter de Gruyter, 1971, no. 2

  16. [24]

    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

  17. [25]

    Neural networks fail to learn periodic functions and how to fix it,

    L. Ziyin, T. Hartwig, and M. Ueda, “Neural networks fail to learn periodic functions and how to fix it,” Advances in Neural Information Processing Systems, vol. 33, pp. 1583–1594, 2020

  18. [26]

    BigV- GAN: A universal neural vocoder with large-scale training,

    S. gil Lee, W. Ping, B. Ginsburg, B. Catanzaro, and S. Yoon, “BigV- GAN: A universal neural vocoder with large-scale training,” in The Eleventh International Conference on Learning Representations , 2023

  19. [27]

    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,” in ICASSP 2021- 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2021, pp. 21–25

  20. [28]

    Parallel wavegan: A fast waveform generation model based on generative adversarial networks with multi-resolution spectrogram,

    R. Yamamoto, E. Song, and J.-M. Kim, “Parallel wavegan: A fast waveform generation model based on generative adversarial networks with multi-resolution spectrogram,” in ICASSP 2020-2020 IEEE In- ternational Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, ...

  21. [29]

    Real time speech enhancement in the waveform domain,

    A. D ´efossez, G. Synnaeve, and Y . Adi, “Real time speech enhancement in the waveform domain,” in Proc. Interspeech 2020 , 2020, pp. 3291– 3295

  22. [30]

    The interspeech 2020 deep noise suppression challenge: Datasets, subjective testing framework, and challenge results,

    C. K. Reddy, V . Gopal, R. Cutler, E. Beyrami, R. Cheng, H. Dubey, S. Matusevych, R. Aichner, A. Aazami, S. Braun, P. Rana, S. Srinivasan, and J. Gehrke, “The interspeech 2020 deep noise suppression challenge: Datasets, subjective testing framework, and challenge results,” in ...

  23. [31]

    Gated residual networks with dilated convolutions for monaural speech enhancement,

    K. Tan, J. Chen, and D. Wang, “Gated residual networks with dilated convolutions for monaural speech enhancement,” IEEE/ACM Transac- tions on Audio, Speech, and Language Processing , vol. 27, no. 1, pp. 189–198, 2019

  24. [32]

    Adam: A method for stochastic optimization,

    D. P. Kingma, “Adam: A method for stochastic optimization,” arXiv preprint arXiv:1412.6980, 2014

  25. [33]

    V ocbench: A neural vocoder benchmark for speech synthesis,

    E. A. AlBadawy, A. Gibiansky, Q. He, J. Wu, M.-C. Chang, and S. Lyu, “V ocbench: A neural vocoder benchmark for speech synthesis,” in ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , 2022, pp. 881–885

  26. [34]

    Per- ceptual evaluation of speech quality (pesq)-a new method for speech quality assessment of telephone networks and codecs,

    A. W. Rix, J. G. Beerends, M. P. Hollier, and A. P. Hekstra, “Per- ceptual evaluation of speech quality (pesq)-a new method for speech quality assessment of telephone networks and codecs,” in 2001 IEEE international conference on acoustics, speech, and signal processing. Proce...

  27. [35]

    An algorithm for intelligibility prediction of time–frequency weighted noisy speech,

    C. H. Taal, R. C. Hendriks, R. Heusdens, and J. Jensen, “An algorithm for intelligibility prediction of time–frequency weighted noisy speech,” IEEE Transactions on audio, speech, and language processing , vol. 19, no. 7, pp. 2125–2136, 2011

  28. [36]

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

    C. K. Reddy, V . Gopal, and R. Cutler, “Dnsmos: A non-intrusive perceptual objective speech quality metric to evaluate noise suppressors,” in ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2021, pp. 6493–6497

  29. [37]

    Optimizing voice conversion network with cycle consistency loss of speaker identity,

    H. Du, X. Tian, L. Xie, and H. Li, “Optimizing voice conversion network with cycle consistency loss of speaker identity,” in 2021 IEEE Spoken Language Technology Workshop (SLT) , 2021, pp. 507–513

  30. [38]

    Image quality assessment: from error visibility to structural similarity,

    Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, “Image quality assessment: from error visibility to structural similarity,” IEEE transactions on image processing , vol. 13, no. 4, pp. 600–612, 2004

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.