Pith. sign in

REVIEW 4 major objections 5 minor 2 cited by

Deep Active Speech Cancellation with Mamba-Masking Network

T0 review · 4 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

Pith's one-line read This paper claims the first deep-learning system to actively cancel both noise and speech, using a Mamba-Masking network and a secondary-path-aware loss.

desk verdict A solid architectural contribution undermined by an internal inconsistency in the real-time causality argument; the cancellation gains may be offline rather than live. read the letter →

arxiv 2502.01185 v2 pith:4HHPAR7X submitted 2025-02-03 cs.SD cs.AIcs.LGeess.ASeess.SP

classification cs.SDcs.AIcs.LGeess.ASeess.SP
keywords activespeechcancellationnoisecontrolMambastate-spacemodelsanti-signalgenerationmulti-bandprocessingsecondary-path-awarelossfeedforwardANCroomimpulseresponsesimulation
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 a deep neural network can perform active speech cancellation (ASC), generating an anti-signal that cancels not only broadband noise but also speech arriving at an error microphone. The proposed architecture, DeepASC, applies a masking mechanism to a Mamba-encoded reference signal, splits the signal into frequency bands, and trains with a loss that compares both the target and prediction after the secondary acoustic path. On simulated feedforward ANC setups, the paper reports up to 7.2 dB lower NMSE than prior deep-learning baselines for noise cancellation and up to 6.2 dB for speech cancellation, with larger gaps when only speech-active segments are scored. If true, this would be the first deep-learning system to cancel speech rather than only noise, with consequences for acoustic privacy, hearing devices, and conversation-aware noise control.

What carries the argument

The central object is the masking interaction between a Mamba state-space encoder and the encoded reference signal. Each frequency band xi is encoded to a latent Hi, a Mamba block Bi emits a mask Mi of the same shape, and the element-wise product Hi·Mi is the masked representation that is concatenated and decoded into the anti-signal; the paper argues this mask learns phase alignment with the reference, which matters for high-frequency speech. A second load-bearing mechanism is the NOAS loss, which projects both the target and the prediction through the secondary path S so the network is supervised in the space where cancellation is actually evaluated, avoiding misleading gradients from frequency-dependent attenuation.

What would settle it

Run the same trained DeepASC model on a physical feedforward ANC testbed with measured primary and secondary paths and a real loudspeaker, comparing time-aligned NMSE at the error microphone against the simulator's prediction; if the 6-7 dB margins over DeepANC and ARN shrink or vanish under measured paths, the central claim fails. A cheaper partial check is to replace the image-source RIRs with the paper's own measured real-world path dataset while keeping all other training details fixed.

Watch

Extended reading notes

Core claim

DeepASC is a feedforward active cancellation network that takes the reference microphone signal x(n), splits it into a full band plus Q sub-bands, encodes each band with a 1-D convolution, and multiplies the encoded representation by a mask produced by Mamba-based layers. The masked band representations are concatenated, fused with a 1x1 convolution, and decoded into the canceling signal y(n). The key training innovation is the Near-Optimal Anti-Signal (NOAS) loss: because the secondary path S(z) attenuates frequencies differently from the primary path P(z), a naive loss NMSE[P*x, S*fLS{y}] penalizes the model even when the pre-propagation anti-signal is optimal. The paper therefore precomputes, for each reference signal, a near-optimal anti-signal y* by gradient descent on NMSE[P*x, S*fLS{ỹ}], then fine-tunes with NMSE[S*fLS{y*}, S*fLS{y}], measuring both sides after the secondary path. With this NOAS target, the paper reports consistent NMSE improvements over the DeepANC and ARN baselines under image-source simulated room acoustics and a saturating loudspeaker model.

Load-bearing premise

The paper's reported gains all come from a simulated acoustic environment with image-source room impulse responses and a Scaled Error Function loudspeaker model, so the central claim depends on those simulated primary and secondary paths behaving like the real acoustic paths and loudspeaker saturation they stand in for.

Editorial extensions

If this is right

  • If the result holds, active noise cancellation systems can be trained end-to-end to suppress speech as well as noise, turning a single reference microphone into a privacy filter for conversations.
  • The NOAS loss can be applied to any feedforward ANC model, since it only replaces the supervision target; the paper's ablation shows it adds consistent gains across model sizes and band configurations.
  • The multi-band variant improves high-frequency cancellation, and adding more bands yields further gains, suggesting that band-split processing is a lever for speech-band cancellation independent of raw parameter count.
  • The small single-band model meets the real-time causality constraint in the simulated setup with lower FLOPs than the baselines, so the approach is compatible with edge deployment if the simulation transfers to physical hardware.
  • Voice-activity-detection-masked scoring shows the advantage over baselines is larger on speech-active segments than on full-signal NMSE, which matters for applications where silence dominates the error metric.

Reading between the lines

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

  • If the masking mechanism is genuinely learning phase alignment rather than a static filter, the same architecture should transfer to moving sources with only light retraining; the paper does not test moving microphones or sources, so that is an open extension.
  • The NOAS target is computed per example by gradient descent, which is an expensive preprocessing step; one could train a small auxiliary network to predict y* directly, eliminating the per-example optimization while keeping the S-projected loss.
  • Because the S-projected loss deliberately ignores frequencies where S has deep nulls, a hybrid system that adds a feedback error microphone could recover residual components that feedforward-only cancellation misses.
  • The paper's VAD analysis suggests standard NMSE understates performance on speech; a perceptually masked listening test on speech-active segments would be a natural next evaluation.
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 DeepASC, a Mamba-Masking network for Active Speech Cancellation (ASC) and Active Noise Cancellation (ANC). The architecture splits the reference signal into frequency bands, encodes each band with convolutional layers, applies Mamba-based masks, and decodes a canceling signal. Training uses an NMSE-based ANC loss followed by a NOAS fine-tuning stage in which per-example gradient descent generates near-optimal anti-signal targets. Experiments in a simulated acoustic environment report NMSE improvements over DeepANC and ARN of up to 7.2 dB for noise and 6.2 dB for speech, with additional evaluations on VAD-masked speech-active segments, real-world measured acoustic paths, ablations, and a runtime comparison. Section 6.4 claims real-time compliance through future-frame prediction, but the paper's own causality budget is inconsistent with the stated look-ahead, and the runtime table does not measure per-frame latency.

Significance. If the reported results hold, DeepASC would be a meaningful empirical advance in deep-learning ANC/ASC, and the NOAS loss is a reasonable supervised-distillation strategy rather than a circular use of test data. The paper includes useful ablations (masking, Mamba vs. LSTM/Transformer, multi-band processing, NOAS), a comparison with reimplemented baselines, and validation on measured acoustic paths. However, the central real-time/causality claim is currently unsupported by the paper's own numbers, and the lack of error bars or significance testing weakens the comparative claims. The significance is therefore conditional on reframing the results as offline simulation and providing a proper streaming-latency evaluation.

major comments (4)
  1. [Section 6.4] The causality constraint is stated as T_ANC < T_p - T_s = 2/343 - 0.5/343 = 0.0043 s, which is about 69 samples at 16 kHz. The same paragraph states that all models rely on future-frame prediction within a 0.01 s (160-sample) window. A 160-sample look-ahead is more than twice the allowed 69-sample budget. This is an internal inconsistency, not a simulator-fidelity issue. Consequently, the NMSE results in Tables 5 and 7 cannot be interpreted as achievable by a causal real-time feedforward ANC system under the stated geometry, and the claim that DeepASC 'meets real-time constraints' is not supported by the reported experimental setup.
  2. [Section 6.4, Table 8] The runtime column reports total processing times (e.g., 0.0136 s for DeepASC) for 3-second clips on an Nvidia H100, not per-frame latencies. Aggregate throughput on a single 3-second clip does not establish that the network can produce each output frame within the 4.3 ms acoustic look-ahead in a streaming, causal implementation. To support the real-time claim, the authors need to report per-frame latency under a streaming/causal inference protocol, including buffering and model warm-up effects, or explicitly relabel the results as offline.
  3. [Tables 5-10] No error bars, confidence intervals, or significance tests accompany any of the NMSE tables. Since DeepANC and ARN are reimplemented in-house, and since some reported margins are small (e.g., the 0.09-0.29 dB NOAS gains in Table 10 and the 1.53 dB margin on WSJ in Table 3), single-run comparisons are insufficient to establish systematic improvement. Please report means and standard deviations over multiple training seeds or evaluation folds, and state the number of runs used for each table.
  4. [Sections 5.2 and 6.3] All performance results are obtained in simulation. Section 6.3 uses measured primary and secondary paths from a database, but the evaluation remains offline and does not include real-time loudspeaker output, hardware latency, or streaming constraints. The paper should state explicitly which claims are about the offline simulated system and which are about physical real-time operation; as written, the abstract and Section 6.4 blur this distinction.
minor comments (5)
  1. [Section 5.1, 5.2, 6.1] There are several typos: 'sources' should be 'sourced' in Section 5.1, 'reverbration' should be 'reverberation' in Section 5.2, and 'respectivly' should be 'respectively' in Section 6.1.
  2. [Section 6.1 and Table 6] The text says the PESQ/STOI results are evaluated with nonlinear distortion η2 = ∞, while Table 6's caption says 'with LS nonlinearity (η = 0.5)'. Please reconcile this contradiction.
  3. [Section 6.4 and Table 4] Table 4 lists DeepASC FLOPs as 2.419 G, but the text in Section 6.4 says the single-band small variant requires 2.862 G FLOPs. Please correct the inconsistency.
  4. [Section 6.2] The VAD-masked NMSE expression 'NMSE[VAD(P ∗ x), VAD(S ∗ {fLS{y})]' has mismatched braces and is missing a closing bracket; please rewrite it using consistent notation, and clarify whether the VAD mask is applied to the primary signal, the anti-signal, or both.
  5. [Tables 2 and 3] Tables 2 and 3 use the same dataset acronyms but report different metrics under different conditions; the captions should state explicitly that Table 2 is VAD-masked NMSE on simulated paths and Table 3 is average NMSE on real-world measured paths, to avoid confusion.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the NOAS target is a per-training-example optimization target and the reported gains are held-out model outputs.

full rationale

The paper's derivation chain is self-contained and does not reduce to its inputs. The NOAS loss constructs y* by minimizing NMSE[P*x, S*fLS{tilde y}] per training example and then trains the network with LNOAS = NMSE[S*fLS{y*}, S*fLS{y}]; this is a standard supervised regression target derived from the simulator, not a fitted parameter renamed as a prediction, and the reported NMSE results in Tables 5 and 7 are produced by the model on held-out test clips rather than by the optimizer that defines y*. The 'near-optimal' label is a descriptive consequence of the optimization defining y*, not a self-referential evaluation claim. Baselines (DeepANC, ARN, FxLMS, THF-FxLMS) come from external work and were reimplemented by the authors, while the simulator and loudspeaker model are cited from independent prior sources; no load-bearing self-citation chain is present. The reader-noted real-time/causality inconsistency (10 ms future-frame window versus the 4.3 ms acoustic look-ahead budget) is a physical correctness risk, not a circularity of the derivation.

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

The central claim rests on the assumption that the simulated acoustic environment (image-source RIRs plus SEF loudspeaker saturation) is representative of real conditions, that P and S are known during training, and that the per-example optimizer produces a valid near-optimal target. No global fitted constants are introduced; the network weights are learned and the per-example targets are computed by optimization.

assumptions (4)
  • domain assumption The image-source method and rir_generator provide a faithful model of room acoustics, and the SEF model captures loudspeaker nonlinearity.
    Invoked in Section 5.2 for simulator setup; the entire evaluation is based on these models.
  • domain assumption The primary and secondary acoustic paths P and S are known exactly during training and evaluation.
    Used in Section 4.2 to compute the NOAS target and the ANC loss; in a real system these paths must be estimated online.
  • domain assumption The feedforward reference microphone provides sufficient lookahead (future-frame prediction of 160 samples at 16 kHz) for causal cancellation of speech.
    Stated in Section 6.4 as the basis for meeting the real-time constraint.
  • ad hoc to paper The per-example gradient descent for y* converges to a near-optimal anti-signal for every training example.
    Required by the NOAS loss (Section 4.2); the paper gives no convergence guarantees or sensitivity analysis.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Deep Active Speech Cancellation with Mamba-Masking Network." pith.science (2026). https://pith.science/paper/4HHPAR7X

@misc{pith2026250201185,
  author       = {Pith},
  title        = {Pith review of: Deep Active Speech Cancellation with Mamba-Masking Network},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4HHPAR7X}},
  note         = {Machine review of arXiv:2502.01185}
}
read the original abstract

We present a novel deep learning network for Active Speech Cancellation (ASC), advancing beyond Active Noise Cancellation (ANC) methods by effectively canceling both noise and speech signals. The proposed Mamba-Masking architecture introduces a masking mechanism that directly interacts with the encoded reference signal, enabling adaptive and precisely aligned anti-signal generation-even under rapidly changing, high-frequency conditions, as commonly found in speech. Complementing this, a multi-band segmentation strategy further improves phase alignment across frequency bands. Additionally, we introduce an optimization-driven loss function that provides near-optimal supervisory signals for anti-signal generation. Experimental results demonstrate substantial performance gains, achieving up to 7.2dB improvement in ANC scenarios and 6.2dB in ASC, significantly outperforming existing methods.

Figures

Figures reproduced from arXiv: 2502.01185 by the authors.

Figure 1
Figure 1. Typical feedforward ANC sys [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. DeepASC Architecture [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Comparison of NMSE (dB) over time for different noise types. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Spectrograms and Power Spectra of Speech Signal (00da010c from WSJ) using Different [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: S-projection importance visualization for NOAS optimization. This behavior is attributed to the properties of convolution with a fixed filter (in this case, S), which does not constitute an isometry and thus fails to preserve distances in the original space. As such, o…
Figure 6
Figure 6. Figure 6: Visualization of VAD masks applied to nine different speech signals from WSJ dataset. [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Toward Optimal ANC: Establishing Mutual Information Lower Bound

    cs.IT 2025-05 reject novelty 3.0 of 10

    A proposed unified lower bound for active noise cancellation error combines a mutual-information term and a spectral-support term, but the information-theoretic term is derived incorrectly and is false.

  2. Active Speech Enhancement: Active Speech Denoising Decliping and Deveraberation

    eess.AS 2025-05 conditional novelty 3.0 of 10

    A Transformer-Mamba model that adds a learned correction signal to degraded speech beats adapted active-noise-control baselines on denoising, dereverberation, and declipping in simulation.

Reference graph

Works this paper leans on

76 extracted references · 16 canonical work pages · cited by 2 Pith papers

  1. [1]

    Process of silencing sound oscillations

    Paul Lueg. Process of silencing sound oscillations. US patent 2043416, 1936

  2. [2]

    Active control of sound

    Philip Arthur Nelson and Stephen J Elliott. Active control of sound. Academic press, 1991

  3. [3]

    Active control of vibration

    Christopher C Fuller, Sharon Elliott, and Philip Arthur Nelson. Active control of vibration. Academic press, 1996

  4. [4]

    Active control of noise and vibration

    Colin H Hansen, Scott D Snyder, Xiaojun Qiu, Laura A Brooks, and Danielle J Moreau. Active control of noise and vibration. E & Fn Spon London, 1997

  5. [5]

    Active noise control: a tutorial review

    Sen M Kuo and Dennis R Morgan. Active noise control: a tutorial review. Proceedings of the IEEE, 87(6):943–973, 1999

  6. [6]

    Deep anc: A deep learning approach to active noise control

    Hao Zhang and DeLiang Wang. Deep anc: A deep learning approach to active noise control. Neural Networks, 141:1–10, 2021

  7. [7]

    Had-anc: A hybrid system comprising an adaptive filter and deep neural networks for active noise control

    JungPhil Park, Jeong-Hwan Choi, Yungyeo Kim, and Joon-Hyuk Chang. Had-anc: A hybrid system comprising an adaptive filter and deep neural networks for active noise control. In Proceedings of the Annual Conference of the International Speech Communication Associa- tion, INTERSPEECH, volume 2023, pages 2513–2517. International Speech Communication Association, 2023

  8. [8]

    Deep learning-based active noise control on construction sites

    Alireza Mostafavi and Young-Jin Cha. Deep learning-based active noise control on construction sites. Automation in Construction, 151:104885, 2023

Show all 76 references
  1. [9]

    Dnoisenet: Deep learning-based feedback active noise control in various noisy environments

    Young-Jin Cha, Alireza Mostafavi, and Sukhpreet S Benipal. Dnoisenet: Deep learning-based feedback active noise control in various noisy environments. Engineering Applications of Artificial Intelligence, 121:105971, 2023

  2. [10]

    Generalized performance of neural network controllers for feedforward active control of nonlinear systems

    Alexander Pike and Jordan Cheer. Generalized performance of neural network controllers for feedforward active control of nonlinear systems. 2023

  3. [11]

    Enhancing active noise control through stacked autoencoders: Training strategies, comparative analysis, and evaluation with practical setup

    Deepali Singh, Rinki Gupta, Arun Kumar, and Rajendar Bahl. Enhancing active noise control through stacked autoencoders: Training strategies, comparative analysis, and evaluation with practical setup. Engineering Applications of Artificial Intelligence, 135:108811, 2024

  4. [12]

    Active adaptive sound control in a duct: A computer simulation

    John C Burgess. Active adaptive sound control in a duct: A computer simulation. The Journal of the Acoustical Society of America, 70(3):715–726, 1981

  5. [13]

    Effect of errors in the plant model on the performance of algorithms for adaptive feedforward control

    CC Boucher, SJ Elliott, and PA Nelson. Effect of errors in the plant model on the performance of algorithms for adaptive feedforward control. In IEE Proceedings F (Radar and Signal Processing), volume 138, pages 313–319. IET, 1991

  6. [14]

    Active mitigation of nonlinear noise processes using a novel filtered-s lms algorithm

    Debi Prasad Das and Ganapati Panda. Active mitigation of nonlinear noise processes using a novel filtered-s lms algorithm. IEEE Transactions on Speech and Audio Processing, 12(3): 313–322, 2004

  7. [15]

    Identification of nonlinear dynamic systems using functional link artificial neural networks

    Jagdish Chandra Patra, Ranendra N Pal, BN Chatterji, and Ganapati Panda. Identification of nonlinear dynamic systems using functional link artificial neural networks. IEEE transactions on systems, man, and cybernetics, part b (cybernetics), 29(2):254–262, 1999

  8. [16]

    Adaptive volterra filters for active control of nonlinear noise processes

    Li Tan and Jean Jiang. Adaptive volterra filters for active control of nonlinear noise processes. IEEE Transactions on signal processing, 49(8):1667–1676, 2001

  9. [17]

    Nonlinear adaptive bilinear filters for active noise control systems

    Sen M Kuo and Hsien-Tsai Wu. Nonlinear adaptive bilinear filters for active noise control systems. IEEE Transactions on Circuits and Systems I: Regular Papers, 52(3):617–624, 2005

  10. [18]

    Leaky-fxlms algorithm: Stochastic analysis for gaussian data and secondary path modeling error

    Orlando José Tobias and Rui Seara. Leaky-fxlms algorithm: Stochastic analysis for gaussian data and secondary path modeling error. IEEE Transactions on speech and audio processing, 13(6):1217–1230, 2005

  11. [19]

    Nonlinear thf-fxlms algorithm for active noise control with loudspeaker nonlinearity

    Sepehr Ghasemi, Raja Kamil, and Mohammad Hamiruce Marhaban. Nonlinear thf-fxlms algorithm for active noise control with loudspeaker nonlinearity. Asian Journal of Control, 18 (2):502–513, 2016. 10

  12. [20]

    Noise cancellation with static mixtures of a nonstationary signal and stationary noise

    Sharon Gannot and Arie Yeredor. Noise cancellation with static mixtures of a nonstationary signal and stationary noise. EURASIP Journal on Advances in Signal Processing, 2002:1–13, 2003

  13. [21]

    Single- sensor active noise cancellation

    Alan V Oppenheim, Ehud Weinstein, Kambiz C Zangi, Meir Feder, and Dan Gauger. Single- sensor active noise cancellation. IEEE Transactions on Speech and Audio Processing, 2(2): 285–290, 1994

  14. [22]

    Kalmannet: Data-driven kalman filtering

    Guy Revach, Nir Shlezinger, Ruud JG Van Sloun, and Yonina C Eldar. Kalmannet: Data-driven kalman filtering. In ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 3905–3909. IEEE, 2021

  15. [23]

    Spherical loudspeaker array for local active control of sound

    Boaz Rafaely. Spherical loudspeaker array for local active control of sound. The Journal of the Acoustical Society of America, 125(5):3006–3017, 2009

  16. [24]

    Feedforward selective fixed-filter active noise control: Algorithm and implementation

    Dongyuan Shi, Woon-Seng Gan, Bhan Lam, and Shulin Wen. Feedforward selective fixed-filter active noise control: Algorithm and implementation. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 28:1479–1492, 2020

  17. [25]

    Selective fixed- filter active noise control based on convolutional neural network

    Dongyuan Shi, Bhan Lam, Kenneth Ooi, Xiaoyi Shen, and Woon-Seng Gan. Selective fixed- filter active noise control based on convolutional neural network. Signal Processing, 190: 108317, 2022

  18. [26]

    A hybrid sfanc-fxnlms algorithm for active noise control based on deep learning

    Zhengding Luo, Dongyuan Shi, and Woon-Seng Gan. A hybrid sfanc-fxnlms algorithm for active noise control based on deep learning. IEEE Signal Processing Letters, 29:1102–1106, 2022

  19. [27]

    Integrated 3d active noise cancellation simulation and synthesis platform using tcl

    Seunghyun Park and Daejin Park. Integrated 3d active noise cancellation simulation and synthesis platform using tcl. In 2023 IEEE 16th International Symposium on Embedded Multicore/Many-core Systems-on-Chip (MCSoC), pages 111–116. IEEE, 2023

  20. [28]

    Transferable latent of cnn-based selective fixed-filter active noise control

    Dongyuan Shi, Woon-Seng Gan, Bhan Lam, Zhengding Luo, and Xiaoyi Shen. Transferable latent of cnn-based selective fixed-filter active noise control. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 31:2910–2921, 2023

  21. [29]

    Deep generative fixed-filter active noise control

    Zhengding Luo, Dongyuan Shi, Xiaoyi Shen, Junwei Ji, and Woon-Seng Gan. Deep generative fixed-filter active noise control. In ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 1–5. IEEE, 2023

  22. [30]

    Delayless generative fixed- filter active noise control based on deep learning and bayesian filter

    Zhengding Luo, Dongyuan Shi, Woon-Seng Gan, and Qirui Huang. Delayless generative fixed- filter active noise control based on deep learning and bayesian filter. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2023

  23. [31]

    Unsupervised learning based end-to-end delayless generative fixed-filter active noise control

    Zhengding Luo, Dongyuan Shi, Xiaoyi Shen, and Woon-Seng Gan. Unsupervised learning based end-to-end delayless generative fixed-filter active noise control. In ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 441–445. IEEE, 2024

  24. [32]

    A new dropout leaky control strategy for multi-channel narrowband active noise cancellation in irregular reverberation room

    Wenzhao Zhu, Bo Xu, Zong Meng, and Lei Luo. A new dropout leaky control strategy for multi-channel narrowband active noise cancellation in irregular reverberation room. In 2021 7th International Conference on Computer and Communications (ICCC), pages 1773–1777. IEEE, 2021

  25. [33]

    Integration of anomaly machine sound detection into active noise control to shape the residual sound

    Chuang Shi, Mengjie Huang, Huitian Jiang, and Huiyong Li. Integration of anomaly machine sound detection into active noise control to shape the residual sound. In ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 8692–869...

  26. [34]

    Deep mcanc: A deep learning approach to multi-channel active noise control

    Hao Zhang and DeLiang Wang. Deep mcanc: A deep learning approach to multi-channel active noise control. Neural Networks, 158:318–327, 2023

  27. [35]

    Multichannel two-gradient direction filtered reference least mean square algorithm for output-constrained multichannel active noise control

    Dongyuan Shi, Bhan Lam, Xiaoyi Shen, and Woon-Seng Gan. Multichannel two-gradient direction filtered reference least mean square algorithm for output-constrained multichannel active noise control. Signal Processing, 207:108938, 2023. 11

  28. [36]

    Remote micro- phone technique for active noise control over distributed networks

    Christian Antoñanzas, Miguel Ferrer, Maria De Diego, and Alberto Gonzalez. Remote micro- phone technique for active noise control over distributed networks. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 31:1522–1535, 2023

  29. [37]

    Spatially selective active noise control systems

    Tong Xiao, Buye Xu, and Chuming Zhao. Spatially selective active noise control systems. The Journal of the Acoustical Society of America, 153(5):2733–2733, 2023

  30. [38]

    A time- domain multi-channel directional active noise control system

    Huawei Zhang, Jihui Zhang, Fei Ma, Prasanga N Samarasinghe, and Huiyuan Sun. A time- domain multi-channel directional active noise control system. In 2023 31st European Signal Processing Conference (EUSIPCO), pages 376–380. IEEE, 2023

  31. [39]

    What is behind the meta-learning initialization of adaptive filter?—a naive method for accelerating convergence of adaptive multichannel active noise control

    Dongyuan Shi, Woon-seng Gan, Xiaoyi Shen, Zhengding Luo, and Junwei Ji. What is behind the meta-learning initialization of adaptive filter?—a naive method for accelerating convergence of adaptive multichannel active noise control. Neural Networks, 172:106145, 2024

  32. [40]

    Gfanc-kalman: Generative fixed-filter active noise control with cnn-kalman filtering

    Zhengding Luo, Dongyuan Shi, Xiaoyi Shen, Junwei Ji, and Woon-Seng Gan. Gfanc-kalman: Generative fixed-filter active noise control with cnn-kalman filtering. IEEE Signal Processing Letters, 2023

  33. [41]

    Low-latency active noise control using attentive recurrent network

    Hao Zhang, Ashutosh Pandey, et al. Low-latency active noise control using attentive recurrent network. IEEE/ACM transactions on audio, speech, and language processing, 31:1114–1123, 2023

  34. [42]

    Self-attending rnn for speech enhancement to improve cross-corpus generalization

    Ashutosh Pandey and DeLiang Wang. Self-attending rnn for speech enhancement to improve cross-corpus generalization. IEEE/ACM Transactions on Audio, Speech, and Language Pro- cessing, 30:1374–1385, 2022

  35. [43]

    An improved artificial bee colony algorithm for model-free active noise control: algorithm and implementation

    Xing Ren and Hongwei Zhang. An improved artificial bee colony algorithm for model-free active noise control: algorithm and implementation. IEEE Transactions on Instrumentation and Measurement, 71:1–11, 2022

  36. [44]

    Genetic algorithm-based adaptive active noise control without secondary path identification

    Yang Zhou, Haiquan Zhao, and Dongxu Liu. Genetic algorithm-based adaptive active noise control without secondary path identification. IEEE Transactions on Instrumentation and Measurement, 2023

  37. [45]

    Speech emission control using active cancellation

    Kazuhiro Kondo and Kiyoshi Nakagawa. Speech emission control using active cancellation. Speech communication, 49(9):687–696, 2007

  38. [46]

    Active speech control using wave- domain processing with a linear wall of dipole secondary sources

    Jacob Donley, Christian Ritz, and W Bastiaan Kleijn. Active speech control using wave- domain processing with a linear wall of dipole secondary sources. In 2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 456–460. IEEE, 2017

  39. [47]

    Computationally efficient fixed-filter anc for speech based on long-term prediction for headphone applications

    Yurii Iotov, Sidsel Marie Nørholm, Valiantsin Belyi, Mads Dyrholm, and Mads Græsbøll Christensen. Computationally efficient fixed-filter anc for speech based on long-term prediction for headphone applications. In ICASSP 2022-2022 IEEE International Conference on Acoustics, Spe...

  40. [48]

    Adaptive sparse linear prediction in fixed-filter anc headphone applications for multi-speaker speech reduction

    Yurii Iotov, Sidsel Marie Nørholm, Valiantsin Belyi, and Mads Græsbøll Christensen. Adaptive sparse linear prediction in fixed-filter anc headphone applications for multi-speaker speech reduction. In 2023 IEEE Workshop on Applications of Signal Processing to Audio and Acoustic...

  41. [49]

    Mamba: Linear-time sequence modeling with selective state spaces

    Albert Gu and Tri Dao. Mamba: Linear-time sequence modeling with selective state spaces. arXiv preprint arXiv:2312.00752, 2023

  42. [50]

    Transformers are ssms: Generalized models and efficient algorithms through structured state space duality

    Tri Dao and Albert Gu. Transformers are ssms: Generalized models and efficient algorithms through structured state space duality. arXiv preprint arXiv:2405.21060, 2024

  43. [51]

    Mamba in speech: Towards an alternative to self-attention

    Xiangyu Zhang, Qiquan Zhang, Hexin Liu, Tianyi Xiao, Xinyuan Qian, Beena Ahmed, Eliathamby Ambikairajah, Haizhou Li, and Julien Epps. Mamba in speech: Towards an alternative to self-attention. arXiv preprint arXiv:2405.12609, 2024

  44. [52]

    Rethinking mamba in speech processing by self-supervised models

    Xiangyu Zhang, Jianbo Ma, Mostafa Shahin, Beena Ahmed, and Julien Epps. Rethinking mamba in speech processing by self-supervised models. arXiv preprint arXiv:2409.07273, 2024. 12

  45. [53]

    Dual-path mamba: Short and long-term bidirectional selective structured state space models for speech separation

    Xilin Jiang, Cong Han, and Nima Mesgarani. Dual-path mamba: Short and long-term bidirectional selective structured state space models for speech separation. arXiv preprint arXiv:2403.18257, 2024

  46. [54]

    Spmamba: State-space model is all you need in speech separation

    Kai Li and Guo Chen. Spmamba: State-space model is all you need in speech separation. arXiv preprint arXiv:2404.02063, 2024

  47. [55]

    An investigation of incorporating mamba for speech enhance- ment

    Rong Chao, Wen-Huang Cheng, Moreno La Quatra, Sabato Marco Siniscalchi, Chao-Han Huck Yang, Szu-Wei Fu, and Yu Tsao. An investigation of incorporating mamba for speech enhance- ment. arXiv preprint arXiv:2405.06573, 2024

  48. [56]

    Mambagan: Mamba based metric gan for monaural speech enhancement

    Tianhao Luo, Feng Zhou, and Zhongxin Bai. Mambagan: Mamba based metric gan for monaural speech enhancement. In 2024 International Conference on Asian Language Processing (IALP), pages 411–416. IEEE, 2024

  49. [57]

    Multichannel long-term streaming neural speech enhance- ment for static and moving speakers

    Changsheng Quan and Xiaofei Li. Multichannel long-term streaming neural speech enhance- ment for static and moving speakers. arXiv preprint arXiv:2403.07675, 2024

  50. [58]

    Wave-u-mamba: An end-to-end framework for high-quality and efficient speech super resolution

    Yongjoon Lee and Chanwoo Kim. Wave-u-mamba: An end-to-end framework for high-quality and efficient speech super resolution. arXiv preprint arXiv:2403.09337, 2024

  51. [59]

    Speech slytherin: Examining the performance and efficiency of mamba for speech separation, recognition, and synthesis

    Xilin Jiang, Yinghao Aaron Li, Adrian Nicolas Florea, Cong Han, and Nima Mesgarani. Speech slytherin: Examining the performance and efficiency of mamba for speech separation, recognition, and synthesis. arXiv preprint arXiv:2407.09732, 2024

  52. [60]

    Ssamba: Self-supervised audio representation learning with mamba state space model

    Siavash Shams, Sukru Samet Dindar, Xilin Jiang, and Nima Mesgarani. Ssamba: Self-supervised audio representation learning with mamba state space model. arXiv preprint arXiv:2405.11831, 2024

  53. [61]

    Audio mamba: Selective state spaces for self-supervised audio representations

    Sarthak Yadav and Zheng-Hua Tan. Audio mamba: Selective state spaces for self-supervised audio representations. arXiv preprint arXiv:2406.02178, 2024

  54. [62]

    Audio mamba: Bidirec- tional state space model for audio representation learning

    Mehmet Hamza Erol, Arda Senocak, Jiu Feng, and Joon Son Chung. Audio mamba: Bidirec- tional state space model for audio representation learning. arXiv preprint arXiv:2406.03344, 2024

  55. [63]

    Tf-mamba: A time-frequency network for sound source localization

    Yang Xiao and Rohan Kumar Das. Tf-mamba: A time-frequency network for sound source localization. arXiv preprint arXiv:2409.05034, 2024

  56. [64]

    Seld-mamba: Selective state-space model for sound event localization and detection with source distance estimation

    Da Mu, Zhicheng Zhang, Haobo Yue, Zehao Wang, Jin Tang, and Jianqin Yin. Seld-mamba: Selective state-space model for sound event localization and detection with source distance estimation. arXiv preprint arXiv:2408.05057, 2024

  57. [65]

    Audio mamba: Pretrained audio state space model for audio tagging

    Jiaju Lin and Haoxuan Hu. Audio mamba: Pretrained audio state space model for audio tagging. arXiv preprint arXiv:2405.13636, 2024

  58. [66]

    Rawbmamba: End-to-end bidirectional state space model for audio deepfake detection

    Yujie Chen, Jiangyan Yi, Jun Xue, Chenglong Wang, Xiaohui Zhang, Shunbo Dong, Siding Zeng, Jianhua Tao, Lv Zhao, and Cunhang Fan. Rawbmamba: End-to-end bidirectional state space model for audio deepfake detection. arXiv preprint arXiv:2406.06086, 2024

  59. [67]

    Audio set: An ontology and human-labeled dataset for audio events

    Jort F Gemmeke, Daniel PW Ellis, Dylan Freedman, Aren Jansen, Wade Lawrence, R Channing Moore, Manoj Plakal, and Marvin Ritter. Audio set: An ontology and human-labeled dataset for audio events. In 2017 IEEE international conference on acoustics, speech and signal processing (...

  60. [68]

    Assessment for automatic speech recognition: Ii

    Andrew Varga and Herman JM Steeneken. Assessment for automatic speech recognition: Ii. noisex-92: A database and an experiment to study the effect of additive noise on speech recognition systems. Speech communication, 12(3):247–251, 1993

  61. [69]

    Timit acoustic phonetic continuous speech corpus.Linguistic Data Consortium, 1993, 1993

    John S Garofolo. Timit acoustic phonetic continuous speech corpus.Linguistic Data Consortium, 1993, 1993

  62. [70]

    Librispeech: an asr corpus based on public domain audio books

    Vassil Panayotov, Guoguo Chen, Daniel Povey, and Sanjeev Khudanpur. Librispeech: an asr corpus based on public domain audio books. In 2015 IEEE international conference on acoustics, speech and signal processing (ICASSP), pages 5206–5210. IEEE, 2015. 13

  63. [71]

    Csr-i (wsj0) complete ldc93s6a

    John Garofolo, David Graff, Doug Paul, and David Pallett. Csr-i (wsj0) complete ldc93s6a. Web Download. Philadelphia: Linguistic Data Consortium, 83, 1993

  64. [72]

    Image method for efficiently simulating small-room acoustics

    Jont B Allen and David A Berkley. Image method for efficiently simulating small-room acoustics. The Journal of the Acoustical Society of America, 65(4):943–950, 1979

  65. [73]

    Room impulse response generator

    Emanuel AP Habets. Room impulse response generator. Technische Universiteit Eindhoven, Tech. Rep, 2(2.4):1, 2006

  66. [74]

    On the lms algorithm with constant and variable leakage factor in a nonlinear environment

    Orlando José Tobias and Rui Seara. On the lms algorithm with constant and variable leakage factor in a nonlinear environment. IEEE transactions on signal processing, 54(9):3448–3458, 2006

  67. [75]

    Adam: A method for stochastic optimization

    P Kingma Diederik. Adam: A method for stochastic optimization. (No Title), 2014

  68. [76]

    + S - Multiband - NOAS

    Stefan Liebich, Johannes Fabry, Peter Jax, and Peter Vary. Acoustic path database for anc in-ear headphone development. 2019. URL https://api.semanticscholar.org/CorpusID: 204793245. 14 Appendix A NOAS Design Choices & Motivation As previously discussed, the optimization proce...

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.