Pith. sign in

REVIEW 4 major objections 5 minor 28 references

SaD: A Scenario-Aware Discriminator for Speech Enhancement

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

Pith's one-line read A scenario-aware discriminator that splits enhanced speech into frequency bands and scores each band separately improves GAN-based speech enhancement without changing the generator.

desk verdict The discriminator swap helps most metrics, but the paper never isolates the frequency split that is its claimed novelty, so the mechanism is underdetermined. read the letter →

arxiv 2509.00405 v2 pith:ORFSLTGE submitted 2025-08-30 cs.SD eess.AS

classification cs.SDeess.AS
keywords speechenhancementgenerativeadversarialnetworkscenario-awarediscriminatorfrequencybandsplittingmetric-basedSNR-drivenlossweightingDNSMOSDFKD
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 argues that in GAN-based speech enhancement, the discriminator's view of the signal is too coarse. Instead of scoring the whole enhanced utterance at once, the proposed SaD module first predicts a frequency division point per utterance, splits the enhanced spectrogram into a low-frequency part where speech dominates and a high-frequency part where noise dominates, and then scores each part with its own quality predictor, plus a full-band score. The split and the loss balance are made scenario-aware: early training borrows division-point labels from the authors' earlier dynamic frequency-division method DFKD, then the labels are removed so the splitter adapts, and the balance between noise-suppression and speech-preservation losses is set by the utterance's signal-to-noise ratio. The paper shows that swapping in this discriminator, without touching the generator, improves PESQ and other perceptual metrics for MetricGAN, CMGAN, and Multi-CMGAN on VoiceBank+DEMAND and DNS2020. The practical payoff is a drop-in component that promises further gains from already-trained or fixed generators.

What carries the argument

The key object is the Scenario-Aware Frequency Splitter (SAFS): a small network that fuses noisy input and enhanced output and predicts one division point m per utterance. The enhanced spectrogram is sliced at m into Y_high and Y_low. Three metric discriminators run in parallel: D1 scores the high-frequency slice against a background-intrusiveness (BAK) target, D2 scores the low-frequency slice against a speech-distortion (SIG) target, and D3 scores the full band against an overall (OVERALL) target. Targets come from a DNSMOS-based quality model retrained for dynamic-length band inputs. Early SAFS training is supervised by DFKD labels; later the labels drop out and the splitter adapts. Signa

What would settle it

Hold the generator fixed and replace SaD's adaptive split with a fixed 4 kHz split; if the PESQ gains over the base generator disappear, the adaptive split is the active ingredient. A second check: run SaD on babble noise, where speech and noise occupy the same frequency band, and see whether the two-band assumption still yields gains.

Watch

Extended reading notes

Core claim

The central claim is that a discriminator which evaluates enhanced speech band-by-band, with the split point and loss balance tailored to the acoustic scene, produces better enhancement than a full-band metric discriminator, and that this holds across generator architectures. On VoiceBank+DEMAND, CMGAN + SaD reaches PESQ 3.622 versus 3.406 for CMGAN alone, with consistent gains in CSIG, CBAK, COVL and modest STOI changes; MetricGAN and Multi-CMGAN also improve. The frequency analysis shows the main visible effect is suppression of spurious high-frequency harmonics that the base CMGAN generates, which improves the perceived listening experience.

Load-bearing premise

The central premise is that one scalar frequency division point per utterance separates speech-dominant from noise-dominant bands well enough for band-wise scoring to help, and that the DFKD-computed division labels used in early training are accurate enough to teach that split.

Editorial extensions

If this is right

  • Because SaD leaves the generator untouched, any future GAN-based enhancement generator can adopt it as a drop-in discriminator upgrade, making its gains stack with generator improvements.
  • Band-specific scoring directs optimization toward two distinct objectives at once—removing high-frequency noise artifacts and preserving low-frequency speech detail—rather than a single blended score.
  • The SNR-driven loss weight removes a manual tuning knob: the model itself decides whether a scene needs more noise suppression or more speech preservation.
  • The adaptive split point handles variability across speakers and noise types better than fixed subband strategies, since the crossover frequency is re-estimated per utterance.
  • The reported CMGAN improvement on VoiceBank+DEMAND (PESQ 3.406 to 3.622) indicates the effect is large enough to matter for downstream applications using perceptual-quality-driven training.

Reading between the lines

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

  • The method implies that discriminator design, not generator capacity, may be the current bottleneck in adversarial speech enhancement—a cheap route to gains without retraining the expensive generator.
  • A natural extension the paper does not develop is predicting multiple split points rather than one, creating more than two bands; this could refine the high-frequency region where noise artifacts concentrate.
  • Because the method's supervision is inherited from DFKD labels and DNSMOS-style scores, its ceiling is tied to those predictors' accuracy; a biased quality model would be amplified rather than corrected by the band-wise training.
  • A concrete stress test would be babble or competing-talker noise, where speech and noise occupy overlapping bands; a single scalar split point is ill-defined there, so the method's assumptions would likely require an overlap-aware reformulation.
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 SaD, a scenario-aware discriminator module that replaces the discriminator in existing GAN-based speech enhancement models without changing the generator. SaD contains a Scenario-Aware Frequency Splitter (SAFS) that predicts a frequency division point from the noisy and enhanced inputs, splits the enhanced spectrum into high- and low-frequency bands, and feeds each band, together with the full-band signal, into three DNSMOS-inspired discriminators trained to predict BAK, SIG, and OVERALL scores. The discriminator loss is reweighted by an SNR-dependent coefficient, and the splitter is initialized with weakly supervised labels computed by the authors' earlier DFKD method. The method is evaluated on MetricGAN, CMGAN, and Multi-CMGAN on VoiceBank+DEMAND and DNS2020, with ablation studies on the weakly supervised label, DNSMOS fine-tuning, and SNR weighting. The central claim is that this plug-in discriminator yields consistent improvements across architectures and datasets.

Significance. If the frequency-split mechanism is genuinely responsible for the reported gains, the proposed module would be a simple, architecture-agnostic improvement for GAN-based speech enhancement, with practical value. The paper has positive features: it evaluates on two public datasets and three representative models; it uses external objective metrics (PESQ, STOI, CSIG, CBAK, COVL); and it includes ablation studies that test several design choices. The reported gains for CMGAN on VoiceBank+DEMAND (PESQ 3.406 to 3.622) are nontrivial. However, the current experiments do not isolate the mechanism claimed to be novel, and several reported results contradict the prose claims of consistency. The contribution is therefore promising but underdetermined as presented.

major comments (4)
  1. [Section 2.3 and Table 3] The paper's central claim is that the scenario-aware frequency split causes the improvements. However, no ablation removes the split while preserving the DNSMOS-based discriminators. Replacing the original discriminator with SaD changes at least four aspects at once: (i) regression targets become DNSMOS BAK/SIG/OVERALL instead of the original metric scores; (ii) three discriminators replace one; (iii) the loss is reweighted by SNR (Eq. 7); and (iv) the enhanced signal is split by SAFS (Eqs. 1-2). The ablations in Table 3 remove the weakly supervised label, DNSMOS fine-tuning, and SNR weighting, but no condition keeps the three DNSMOS-based discriminators and simply removes the frequency split. A no-split control—same discriminator architecture and DNSMOS losses on full-band input—is required to attribute the gains to the SAFS. Without it, the results are consistent with the weaker explan
  2. [Section 3.3.1, Tables 1-2] The prose overstates consistency. In Table 2, MetricGAN+SaD is worse than MetricGAN on CSIG (3.880 vs 3.903), CBAK (2.457 vs 2.516), and STOI (0.894 vs 0.912). In Table 1, STOI drops for MetricGAN (0.876 to 0.868) and CMGAN (0.958 to 0.947). In Table 3, the variant without SNR weighting has higher STOI (0.951 vs 0.947), and the variant without DNSMOS fine-tuning has higher CBAK (3.328 vs 3.24) than the full model. The statements that the method shows 'consistent improvements across all three models' on DNS2020 and 'nearly all metrics' on VoiceBank+DEMAND are therefore not supported by the tables. Additionally, no error bars or significance tests are reported, so it is impossible to judge whether even the headline PESQ gains are reliable. Please report variance or significance and temper the consistency claims.
  3. [Section 2.2 and Eq. (3)] The weakly supervised label m is computed by DFKD [18], a prior paper with overlapping authorship, and this manuscript provides no independent validation of those labels on VoiceBank+DEMAND or DNS2020. Table 3 shows that removing this supervision degrades PESQ from 3.622 to 3.539 and STOI from 0.947 to 0.849. Since the SAFS is initialized and guided by these labels, and since the frequency split is the paper's main novelty, the dependence on this self-cited, unvalidated label source is load-bearing. The authors should either validate the DFKD labels (e.g., by comparison with spectrographic speech-dominance boundaries or perceptual judgments) or show that the converged unsupervised splitter produces sensible, scenario-dependent division points. Otherwise, the observed improvements may be attributable to the DFKD prior rather than to learned scenario-awareness.
  4. [Section 3.3.1, Table 2 (MetricGAN row)] The DNS2020 result for MetricGAN+SaD is not supportive of the generalization claim: PESQ improves marginally (2.647 to 2.663), while CSIG, CBAK, and STOI degrade. This is the only case where the method is applied to a non-conformer generator (BLSTM), so the claim that SaD 'can effectively adapt to various generator architectures' rests substantially on this row. The authors should either explain why the frequency-split mechanism fails to help for MetricGAN on DNS2020, or explicitly restrict the generalization claim to conformer-based generators.
minor comments (5)
  1. [Section 1] The introduction states that 'Section 5 presents concluding remarks', but the conclusions appear in Section 4. Please fix the cross-reference.
  2. [Eq. (2)] The notation \hat{Y}[:\hat{m}] and \hat{Y}[\hat{m}:] is not defined. Specify whether \hat{Y} is the STFT magnitude or complex spectrum, and define \hat{m} in frequency bins or Hz.
  3. [Eq. (7)] The definition \alpha = SNR/SNR_{max} is problematic when SNR is negative or zero. State how SNR is computed and whether it is clipped or shifted before computing the weight.
  4. [Table 3] The label 'w/o DNSMOS fine-tune' is ambiguous. It could mean removing the pre-training of the discriminators on DNSMOS-labeled data, or replacing the DNSMOS targets with the original metric scores. Please clarify exactly which component is removed.
  5. [Figure 1 and Section 2.3] Figure 1 labels the discriminators as 'pre-trained metric estimation discriminators', while Section 2.3 describes retraining them to handle dynamic band lengths. Harmonize the terminology.

Circularity Check

1 steps flagged · score 4.0 of 10

Partial circularity: the frequency-split supervision is a self-cited, load-bearing input; the central PESQ gains are externally measured, so the derivation is not a tautology.

  1. self citation load bearing [Section 2.2, Eq. (3); Section 3.3.2 Table 3; Ref. [18]]
    "Specifically, we adopt the method of DFKD to calculate the frequency division point of the clean speech as the label for frequency division and utilize this label to guide the initial training phase of the SaD-GAN network. ... Lossm = ∥m − mˆ∥2 (3). The paper itself also notes DFKD 'relies heavily on empirical computation, which may not always yield optimal results' (Section 1)."

    The supervised target m in Eq. (3) is produced by DFKD, a prior work by the same authors (Ref. [18]). The SAFS is trained to regress exactly this self-cited quantity, so the 'scenario-aware' frequency split is, by construction, a learned imitation of a heuristic imported from an overlapping-authors paper. The only ablation that removes this supervision ('w/o weakly supervised') lowers PESQ from 3.622 to 3.539, showing the headline gain depends on the validity of that self-cited label. The paper does not independently validate DFKD's division points against any ground truth, and even concedes they 'may not always yield optimal results.' This makes the frequency-splitting mechanism load-bearing on a self-citation, although the final evaluation metrics (PESQ, STOI, CSIG, CBAK, COVL) are exter

full rationale

The derivation chain from SaD to the reported quality gains is not formally circular: equations (1)-(7) define a training objective, and PESQ/STOI/CSIG/CBAK/COVL are external metrics not used as optimization targets. The main circularity concern is the weakly supervised frequency division: Eq. (3) trains the SAFS to regress m computed by DFKD, a prior method by the same authors, and the ablation in Table 3 shows removing this supervision degrades the main result (PESQ 3.622 to 3.539). Thus the 'scenario-aware' split mechanism depends on an unverified, self-cited heuristic. In addition, no ablation removes the frequency split while keeping the other simultaneous changes (DNSMOS targets, three discriminators, SNR reweighting), so the headline attribution to frequency splitting is underdetermined. However, that is an experimental-confound issue rather than a derivation-level circularity. Because the central empirical claim is still checked against external benchmarks and is not built from the paper's own equations, the appropriate score is moderate partial self-citation dependency, not full circularity.

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

The method rests on several assumptions about spectral structure (speech below 4 kHz), the validity of self-cited DFKD labels, and the transferability of DNSMOS scores to band-limited inputs; none are independently verified. Free parameters (gamma, 10-epoch schedule, SNR_max) are chosen by hand without sensitivity analysis.

free parameters (3)
  • gamma
    Equation (8): weight of discriminator loss in total loss; value not reported.
  • supervised_epochs = 10
    Section 3.2: SAFS uses DFKD labels for first 10 epochs; chosen by hand, no sensitivity analysis.
  • SNR_max = dataset-specific (e.g., 20 dB for DNS2020)
    Equation (7): alpha = SNR/SNR_max; normalization is dataset-dependent and no value is reported.
assumptions (4)
  • domain assumption Speech is the dominant component below 4 kHz and noise dominates above 4 kHz
    Introduction states human speech is concentrated in 1-4 kHz and uses this to justify separating high and low frequency bands for quality scoring.
  • domain assumption DFKD-derived division points are valid ground-truth labels for the frequency split
    Section 2.2: ground-truth label m is computed with DFKD [18], a prior paper by overlapping authors; no independent validation is given, and the ablation shows performance degrades without this supervision.
  • domain assumption DNSMOS BAK/SIG/OVERALL scores computed on full-band clean speech transfer to band-limited, variable-length inputs
    Section 2.3: D1 and D2 are trained on DNSMOS scores of clean speech then applied to high/low parts; no evidence DNSMOS is calibrated for band-limited input.
  • domain assumption A single scalar division point per utterance captures the scenario's frequency characteristics
    Equation (2) splits the entire utterance at m, ignoring that the speech/noise spectral balance changes over time.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SaD: A Scenario-Aware Discriminator for Speech Enhancement." pith.science (2026). https://pith.science/paper/ORFSLTGE

@misc{pith2026250900405,
  author       = {Pith},
  title        = {Pith review of: SaD: A Scenario-Aware Discriminator for Speech Enhancement},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ORFSLTGE}},
  note         = {Machine review of arXiv:2509.00405}
}
read the original abstract

Generative adversarial network-based models have shown remarkable performance in the field of speech enhancement. However, the current optimization strategies for these models predominantly focus on refining the architecture of the generator or enhancing the quality evaluation metrics of the discriminator. This approach often overlooks the rich contextual information inherent in diverse scenarios. In this paper, we propose a scenario-aware discriminator that captures scene-specific features and performs frequency-domain division, thereby enabling a more accurate quality assessment of the enhanced speech generated by the generator. We conducted comprehensive experiments on three representative models using two publicly available datasets. The results demonstrate that our method can effectively adapt to various generator architectures without altering their structure, thereby unlocking further performance gains in speech enhancement across different scenarios.

Figures

Figures reproduced from arXiv: 2509.00405 by the authors.

Figure 1
Figure 1. Overview of SaD. The Scenario-Aware Frequency Splitter receives the enhanced speech generated by the Gen￾erator and the original noisy speech as inputs, and predicts the frequency division points to partition the enhanced speech into high-frequency and low-frequency components. Three dis￾tinct pre-trained metric estimation discriminators are employed to evaluate the quality of the high-frequency component, low￾frequ… view at source ↗
Figure 2
Figure 2. Frequency Analysis for Diverse Acoustic Scenarios. The upper and lower rows depict the time-frequency represen￾tations of two distinct scenarios. From left to right: the leftmost plot illustrates the input noisy signal, the middle plot shows the enhancement result obtained using the official CMGAN open￾source code, and the rightmost plot presents the enhanced re￾sult achieved with CMGAN + SaD. assessment of the mode… view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

28 extracted references · 19 canonical work pages

  1. [18]

    Finally: fast and universal speech enhancement with studio-like quality,

    N. Babaev, K. Tamogashev, A. Saginbaev, I. Shchekotov, H. Bae, H. Sung, W. Lee, H.-Y . Cho, and P. Andreev, “Finally: fast and universal speech enhancement with studio-like quality,”Advances in Neural Information Processing Systems, vol. 37, pp. 934–965, 2025

  2. [1]

    Introduction Speech enhancement (SE) is of paramount importance in mod- ern communication systems and has garnered significant atten- tion due to its applications in various fields such as telecommu- nications, hearing aids, and speech recognition. The advent of deep learning has revolutionized SE, with deep neural network (DNN)-based approaches [1–4] con...

  3. [2]

    System Overview Our proposed method reconfigures the discriminator for SE GAN-like models (e.g., MetricGAN, CMGAN)

    Methodology 2.1. System Overview Our proposed method reconfigures the discriminator for SE GAN-like models (e.g., MetricGAN, CMGAN). Specifically, we introduce a scenario-aware frequency splitter (SAFS) that adaptively partitions the enhanced speech generated by the gen- erator into high-frequency and low-frequency components. The quality of these two fre...

  4. [3]

    Datasets To validate the effectiveness of our proposed methods, we con- ducted experiments using the DNS2020 challenge dataset [20] and the V oiceBank+DEMAND dataset [21]

    Experiments 3.1. Datasets To validate the effectiveness of our proposed methods, we con- ducted experiments using the DNS2020 challenge dataset [20] and the V oiceBank+DEMAND dataset [21]. The dataset employed in the DNS2020 challenge comprises 500 hours of pristine speech recordings from 2,150 unique Table 1: Performance comparison on VoiceBANK+DEMAND. †...

  5. [4]

    Similarly, omitting the pre-fine-tuning of DNSMOS for dynamic inputs led to a no- table decline in most metrics, with the exception of CBAK

    When the SAFS was trained entirely in an unsupervised man- ner throughout the training process, the PESQ, CSIG, CBAK, COVL, and STOI metrics exhibited decreases of 0.083, 0.013, 0.137, 0.065, and 0.098, respectively. Similarly, omitting the pre-fine-tuning of DNSMOS for dynamic inputs led to a no- table decline in most metrics, with the exception of CBAK....

  6. [5]

    Conclusions This study introduces a scenario-aware discriminator tailored for SE models based on the GAN framework. Our approach in- tegrates the time-frequency characteristics of the current acous- tic scenario, partitions the enhanced speech generated by the generator into high- and low-frequency bands, and employs dis- tinct quality evaluation metrics ...

  7. [6]

    Dccrn: Deep complex convolution re- current network for phase-aware speech enhancement,

    Y . Hu, Y . Liu, S. Lv, M. Xing, S. Zhang, Y . Fu, J. Wu, B. Zhang, and L. Xie, “Dccrn: Deep complex convolution re- current network for phase-aware speech enhancement,” arXiv preprint arXiv:2008.00264, 2020

  8. [7]

    A convolutional recurrent neural network for real-time speech enhancement

    K. Tan and D. Wang, “A convolutional recurrent neural network for real-time speech enhancement.” in Interspeech, vol. 2018, 2018, pp. 3229–3233

Show all 28 references
  1. [8]

    Demucs: Deep extractor for music sources with extra unlabeled data remixed,

    A. D ´efossez, N. Usunier, L. Bottou, and F. Bach, “Demucs: Deep extractor for music sources with extra unlabeled data remixed,” arXiv preprint arXiv:1909.01174, 2019

  2. [9]

    Low-latency single channel speech enhancement using u-net convolutional neural networks,

    A. E. Bulut and K. Koishida, “Low-latency single channel speech enhancement using u-net convolutional neural networks,” in ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2020, pp. 6214– 6218

  3. [10]

    Suppression of acoustic noise in speech using spectral subtraction,

    S. Boll, “Suppression of acoustic noise in speech using spectral subtraction,” IEEE Transactions on acoustics, speech, and signal processing, vol. 27, no. 2, pp. 113–120, 1979

  4. [11]

    All-pole modeling of degraded speech,

    J. Lim and A. Oppenheim, “All-pole modeling of degraded speech,” IEEE Transactions on Acoustics, Speech, and Signal Processing, vol. 26, no. 3, pp. 197–210, 1978

  5. [12]

    Segan: Speech enhancement generative adversarial network,

    S. Pascual, A. Bonafonte, and J. Serra, “Segan: Speech enhancement generative adversarial network,” arXiv preprint arXiv:1703.09452, 2017

  6. [13]

    Metricgan: Gen- erative adversarial networks based black-box metric scores opti- mization for speech enhancement,

    S.-W. Fu, C.-F. Liao, Y . Tsao, and S.-D. Lin, “Metricgan: Gen- erative adversarial networks based black-box metric scores opti- mization for speech enhancement,” in International Conference on Machine Learning. PmLR, 2019, pp. 2031–2041

  7. [14]

    Hifi-gan: High-fidelity denois- ing and dereverberation based on speech deep features in adver- sarial networks,

    J. Su, Z. Jin, and A. Finkelstein, “Hifi-gan: High-fidelity denois- ing and dereverberation based on speech deep features in adver- sarial networks,” arXiv preprint arXiv:2006.05694, 2020

  8. [15]

    Cmgan: Conformer- based metric gan for speech enhancement,

    R. Cao, S. Abdulatif, and B. Yang, “Cmgan: Conformer- based metric gan for speech enhancement,” arXiv preprint arXiv:2203.15149, 2022

  9. [16]

    Multi- cmgan+/+: Leveraging multi-objective speech quality metric pre- diction for speech enhancement,

    G. Close, W. Ravenscroft, T. Hain, and S. Goetze, “Multi- cmgan+/+: Leveraging multi-objective speech quality metric pre- diction for speech enhancement,” in ICASSP 2024-2024 IEEE In- ternational Conference on Acoustics, Speech and Signal Process- ing (ICASSP). IEEE, 2024, pp. 351–355

  10. [17]

    Scp-gan: Self- correcting discriminator optimization for training consistency pre- serving metric gan on speech enhancement tasks,

    V . Zadorozhnyy, Q. Ye, and K. Koishida, “Scp-gan: Self- correcting discriminator optimization for training consistency pre- serving metric gan on speech enhancement tasks,” arXiv preprint arXiv:2210.14474, 2022

  11. [19]

    Perceptual 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, “Perceptual 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. Proceed...

  12. [20]

    An al- gorithm for intelligibility prediction of time–frequency weighted noisy speech,

    C. H. Taal, R. C. Hendriks, R. Heusdens, and J. Jensen, “An al- gorithm 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

  13. [21]

    Metricgan+: An improved version of metricgan 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 metricgan for speech enhancement,” arXiv preprint arXiv:2104.03538, 2021

  14. [22]

    Sub-band knowledge distillation framework for speech enhancement,

    X. Hao, S. Wen, X. Su, Y . Liu, G. Gao, and X. Li, “Sub-band knowledge distillation framework for speech enhancement,”arXiv preprint arXiv:2005.14435, 2020

  15. [23]

    Dynamic frequency-adaptive knowledge distillation for speech enhance- ment,

    X. Yuan, S. Liu, H. Chen, L. Zhou, J. Li, and J. Hu, “Dynamic frequency-adaptive knowledge distillation for speech enhance- ment,” arXiv preprint arXiv:2502.04711, 2025

  16. [24]

    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 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2021, pp. 6493–6497

  17. [25]

    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 et al., “The interspeech 2020 deep noise suppression challenge: Datasets, subjective testing framework, and challenge results,” arXiv preprint arXiv:2005.13981, 2020

  18. [26]

    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

  19. [27]

    The diverse environments multi-channel acoustic noise database (demand): A database of multichannel environmental noise recordings,

    J. Thiemann, N. Ito, and E. Vincent, “The diverse environments multi-channel acoustic noise database (demand): A database of multichannel environmental noise recordings,” in Proceedings of Meetings on Acoustics, vol. 19, no. 1, 2013

  20. [28]

    Evaluation of objective quality measures for speech enhancement,

    Y . Hu and P. C. Loizou, “Evaluation of objective quality measures for speech enhancement,” IEEE Transactions on audio, speech, and language processing, vol. 16, no. 1, pp. 229–238, 2007

Pith tools

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