WTFormer: A Wavelet Conformer Network for MIMO Speech Enhancement with Spatial Cues Peservation

Junqi Zhao; Lu Han; Renhua Peng

REVIEW 4 major objections 5 minor 39 references

WTFormer: A Wavelet Conformer Network for MIMO Speech Enhancement with Spatial Cues Peservation

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read WTFormer, a 0.98M-parameter wavelet-conformer MIMO network, matches advanced denoising quality while preserving inter-microphone spatial cues better than larger baselines.

desk verdict A credible, parameter-efficient MIMO enhancement system with a genuinely new combination of components; the weakest point is that the MUSIC-based spatial loss is anchored to the noisy mixture, so the headline spatial-cue gains need stronger validation. read the letter →

arxiv 2506.22001 v1 pith:ORYOYK2S submitted 2025-06-27 eess.AS cs.SD

Lu Han , Junqi Zhao , Renhua Peng This is my paper · ORCID

classification eess.AScs.SD

keywords multichannelspeechenhancementMIMOspatialcuepreservationwaveletconvolutionConformerMUSICspectrummicrophonearrayparameterefficiency

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 show that MIMO speech enhancement—enhancing all microphone channels while keeping their spatial relationships intact—does not require a large network or a beamforming front end. The proposed WTFormer uses wavelet convolutions and a Conformer-based time-frequency module, and it is trained with a MUSIC-spectrum loss to preserve spatial cues. On LibriSpeech, it reaches the highest PESQ among the compared systems, matches EaBNet on STOI and eSTOI, and reports lower inter-microphone ITD, IPD, and ILD errors than MIMO-UNet, with only 0.98M parameters. If this holds, small MIMO networks could serve as spatial-cue-preserving front ends for downstream beamforming and sound-source localization.

What carries the argument

The three load-bearing components are: WTConv, a Haar-wavelet multi-resolution convolution that enlarges the receptive field without adding many parameters; TF-Conformer, two cascaded Conformer blocks that model time-frequency dependencies along with channel correlations; and MCA, a three-branch attention module over spatial, time, and frequency dimensions that replaces skip connections. The training signal is the multi-task loss in Eq. (4), which combines an SI-SNR loss for noise suppression with an MSE between the MUSIC spatial spectrum of the multi-channel input and that of the enhanced output, weighted by two learnable uncertainty parameters. This loss is what pushes the network to keep the inter-channel phase and magnitude relationships that downstream beamforming and DOA estimation rely on.

What would settle it

Compute $\Delta$ITD, $\Delta$IPD, and $\Delta$ILD of WTFormer outputs against the clean early-reverberant target, not the noisy input, across the -5 to 5 dB test SNR range; if the advantages over MIMO-UNet shrink or reverse at low SNR, the MUSIC-proxy training is preserving noise-dominated rather than target spatial structure.

Watch

Extended reading notes

Core claim

The paper claims that a MIMO speech enhancement network built from wavelet convolution blocks, TF-Conformer, and multidimensional collaborative attention, trained with a multi-task loss that includes an MSE on MUSIC spatial spectra, can simultaneously match the denoising performance of a larger modern beamforming network and preserve inter-microphone spatial cues better than MIMO-UNet. On LibriSpeech with a 4-cm-spaced 8-element uniform linear array, WTFormer reaches PESQ 3.02, STOI 0.92, eSTOI 0.84, and SI-SNR 10.31 dB with 0.98M parameters, and it reduces $\Delta$ITD by 27.3% and $\Delta$ILD by 18.0% relative to MIMO-UNet, with $\Delta$IPD 0.75 rad and $\Delta$ILD 0.73 dB.

Load-bearing premise

The training loss rewards matching the spatial pattern of the noisy input, not the clean speech; if that pattern is dominated by noise, the claimed preservation of sound-direction cues may not hold.

Editorial extensions

If this is right

A MIMO enhancement model can match a modern all-neural beamforming network (EaBNet) in PESQ, STOI, and eSTOI while using roughly one-third of the parameters, so spatial-cue preservation does not have to be bought at the cost of denoising quality.
Using the MUSIC spatial spectrum as a training target reduces $\Delta$ITD by 27.3% and $\Delta$ILD by 18.0% compared to MIMO-UNet, suggesting the loss is effective at retaining inter-channel phase and level relationships.
Ablation results show that removing WTConv or MCA degrades both PESQ and $\Delta$ITD, indicating that multi-resolution convolution and multi-dimensional attention are jointly responsible for the gain.
Because the mask generator estimates a complex ideal ratio mask per channel, the enhanced multi-channel output can be fed directly into later stages such as DOA estimation or beamforming without an extra filter-and-sum step.

Reading between the lines

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

The spatial-preservation claim is likely optimistic in low-SNR regimes because the MUSIC loss uses the noisy mixture as the reference; comparing against the clean early-reverberant target would be a stronger test.
The same WTConv plus MCA plus Conformer recipe could transfer to binaural cue preservation or to non-linear array geometries, since the loss operates on the MUSIC spectrum rather than on a specific array configuration.
The 0.98M parameter count suggests wavelet convolution may substitute for deeper encoder stacks in other multichannel audio tasks, not only enhancement, though that generalization is outside the paper's evidence.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

You should know this paper is worth a look if you care about MIMO or binaural enhancement. It is not a paradigm shift, but it is a solid engineering result: a 0.98M-parameter MIMO enhancer that matches EaBNet on PESQ/STOI/eSTOI and reports clearly better ΔITD/ΔILD on a simulated LibriSpeech array test set. The combination of wavelet convolution blocks, a TF-Conformer backbone, multi-dimensional attention for skip connections, and a MUSIC-spectrum loss is new, and the small parameter count makes it practically interesting.

The paper does several things well. The ablation study shows each component contributes. The main comparison is against reasonable baselines, including a causal beamformer and a strong all-neural EaBNet. The spatial-cue evaluation is standard (ΔITD, ΔIPD, ΔILD against the early-reverberant target), and the reported numbers are large enough that I doubt they are pure noise.

Now the soft spots, in proportion. First, no error bars or significance tests anywhere. With a single train/test split, the headline 0.03 PESQ gap over EaBNet and the spatial gains could shift substantially. Second, the EaBNet baseline is described as “filter-and-sum removed for MIMO comparison,” but no detail is given on how that adaptation was done; MIMO-UNet appears to be the original, but EaBNet's architecture may not port cleanly to eight-channel output. Third, and most important, the MUSIC loss in Eq. 4 minimizes the MSE between the enhanced output's spatial spectrum and that of the noisy input Y, not the clean early-reverberant target. The evaluation metrics are computed against the target. In the -5 to 5 dB test SNR range, MUSIC peaks of Y can be noise-dominated, so the loss could be preserving mixture spatial structure rather than target cues. The paper gives no SNR-stratified spatial metrics and no comparison to a target-referenced loss. The ablation only proves the loss matters, not that it points the right way.

The MUSIC implementation itself is also under-specified: 300 narrowband signals against an STFT with 161 bins, no steering-vector model, no source count, no mention of a differentiable eigendecomposition. That is a reproducibility gap, but not a fatal one.

Who is this for? Researchers working on microphone-array preprocessing for hearing aids or binaural listening. It deserves a serious referee, but a referee should push for code, error bars, the EaBNet MIMO baseline detail, and an SNR-stratified analysis of the spatial metrics. My take is conditional acceptance at best, with the spatial-loss reference issue front and center.

Referee Report

4 major / 5 minor

Summary. The paper proposes WTFormer, a multi-channel MIMO speech enhancement network with only 0.98M parameters. The architecture combines wavelet convolution blocks (WTConv), a TF-Conformer module, and multidimensional collaborative attention (MCA), and it estimates multi-channel complex masks to reconstruct all microphone signals. A multi-task loss (Eq. 4) combines an SI-SNR-based noise suppression term with a spatial preservation term Lps, defined as the MSE between the MUSIC spatial spectrum of the multi-channel signal before and after processing. Experiments on simulated LibriSpeech data (8-channel ULA, image-method RIRs) compare WTFormer against Ti-MVDR, MB-MVDR, MIMO-UNet, and EaBNet. Results in Table 2 show WTFormer achieving the highest PESQ, STOI/eSTOI comparable to EaBNet, and lower ΔITD/ΔIPD/ΔILD, along with an ablation study in Table 1.

Significance. If the reported results hold, the paper makes a practically interesting contribution: a very compact MIMO network that matches or exceeds a larger EaBNet in quality metrics while substantially improving spatial cue preservation. The use of a MUSIC-spectrum loss as a spatial regularization is a plausible idea, and the ablation study indicates that the WTConv, MCA, and Lps components each affect the outcome. The paper gives credit for explicit uncertainty-weighted loss balancing (Eq. 4, following Kendall et al.) and for reporting parameter counts for all systems. However, the load-bearing spatial claim rests on the Lps loss being anchored to the noisy input rather than the target speech, and the implementation details of the MUSIC loss are insufficient for reproduction. The absence of error bars or significance tests also makes the magnitudes of the reported spatial improvements uncertain.

major comments (4)

[4.3, Eq. (4)] The spatial-preservation loss Lps is defined as the MSE between the MUSIC spatial spectrum of the multi-channel signal before and after processing, where the 'before' signal is the noisy mixture Y. In contrast, the evaluation metrics in Section 4.4 (ΔITD, ΔIPD, ΔILD) are computed with respect to the early-reverberant target. At the test SNRs of -5 to 5 dB, the MUSIC spectrum of Y may be dominated by noise directions, so minimizing the distance to Y's spectrum may preserve noise-dominated spatial structure rather than the target-speech cues that the metrics measure. The ablation in Table 1 shows that Lps changes the solution, but it does not establish that the noisy reference is the correct anchor. Please provide SNR-stratified spatial metrics, compare Lps against a target-referenced MUSIC loss, or analyze the spectral agreement in speech-dominant versus noise-dominant frequency bands.
[4.3] The MUSIC loss implementation is under-specified. The text states that the signal is divided into 300 narrowband signals and a 300×181 spatial spectrum is obtained, but the source-count assumption, the steering-vector model, the sub-band grouping method, and the way the eigendecomposition is made differentiable are not described. Moreover, Section 4.2.2 uses F=161 frequency bins after STFT, so the relationship between 300 bands and 161 bins is unclear. Without these details, the training objective cannot be reproduced, and the causal link between Lps and the reported spatial improvements cannot be independently verified.
[4.4, Table 2] The adaptation of EaBNet to MIMO evaluation is not validated. The paper states that 'the filter-and-sum step is removed for MIMO comparison,' but EaBNet is a MISO beamforming network trained with a single output channel; simply removing filter-and-sum after training may not yield a valid multi-channel enhancement system. No evidence (e.g., training curves, multi-channel output quality, or channel-wise metrics) is given that this adapted EaBNet is a fair baseline. Please either validate the MIMO adaptation or train a properly configured MIMO variant of EaBNet.
[Tables 1 and 2] All metrics are reported as single values without error bars or significance tests. The differences that support the main spatial claim, such as 84.27 µs versus 104.39 µs in Table 1 and 84.27 µs versus 115.93 µs in Table 2 for ΔITD, are small in absolute terms, and without variance estimates the robustness of these differences is uncertain. Please report results over multiple runs with confidence intervals or at least significance tests.

minor comments (5)

[Title] The title contains a typo: 'Peservation' should be 'Preservation'.
[Introduction] There are typos in the Introduction, including 'Nerual' for 'Neural' and 'signal-noise ration' for 'signal-noise ratio'.
[Eq. (4)] Equation (4) is garbled in the manuscript: the coefficients appear as '10 2σ2 1' and '1 2σ2 1'; these should be 10/(2σ1^2) and 1/(2σ2^2) respectively.
[4.2.1] The sentence describing the Conv2d kernel sizes reads 'with stride (2, 1) in the and frequency time axes'; it should be 'in the time and frequency axes'.
[4.4] The microphone pairs used for ΔITD/ΔIPD/ΔILD are {1,5}, {2,6}, {3,7}, and {4,8}; with a 4 cm inter-element spacing, these pairs span 16 cm, which is a much wider baseline than adjacent microphones. Please clarify whether this selection is intentional for increased cue magnitudes.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's spatial-cue and denoising claims are empirical comparisons against external baselines and target-referenced metrics, not derivations from their own definitions.

full rationale

The paper's central claims are empirical benchmark results, not derived predictions. WTFormer's denoising performance is measured with PESQ, STOI, eSTOI, and SI-SNR against external baseline systems, and its spatial-cue preservation is measured with ΔITD, ΔIPD, and ΔILD computed against the early-reverberant target signal. The only potentially circular-looking element is the MUSIC-based spatial loss Lps in Eq. 4, which minimizes the MSE between the MUSIC spatial spectrum of the enhanced signal and that of the noisy input. But this is not equivalent by construction to the evaluation metrics: Lps anchors to the noisy mixture, whereas ΔITD/ΔIPD/ΔILD are computed after subtracting target values, and the loss-to-metric link is an empirical hypothesis tested by the ablation in Table 1 rather than a definitional identity. The architectural components (WTConv, TF-Conformer, MCA) are attributed to prior external works, not to the authors' own unverified self-citations, and no uniqueness theorem is invoked to force the design. Whether the noisy-anchored loss is the right training objective at low SNR is a correctness or generalization concern, not circularity. The paper is self-contained as an empirical comparison, so the circularity score is 0.

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

The central empirical claim rests on the simulated-data setup, the choice of spatial loss, and the unvalidated modification of the EaBNet baseline. The learnable loss weights and several hand-chosen discretizations are free parameters. No new physical entities are introduced.

free parameters (4)

Loss-balancing weights sigma1, sigma2 = learned, final values not reported
Eq. 4 uses learnable uncertainty weights to balance the SI-SNR enhancement loss and the MUSIC spatial loss; the reported results depend on the learned balance.
MUSIC loss frequency bands and DOA grid = 300 bands, 181 DOA bins, 4 s segment
Section 4.3: these hand-chosen discretizations define the spatial spectrum loss and therefore the spatial preservation behavior.
Early/late reverberation cutoff = not specified
Section 2 defines early-reverberation speech as the training target but never states the RIR cutoff time separating early and late components, which changes the enhancement target.
Room and array simulation configuration = 8-element ULA, 4 cm spacing, room 5-10 m x 3-4 m, RT60 0.3-0.7 s, SNR -5 to 20 dB train, -5 to 5 dB test
Section 4.1: the entire evaluation domain is defined by these hand-chosen simulation parameters; results may not transfer to other geometries or real recordings.

assumptions (5)

domain assumption Haar wavelet convolution preserves inter-channel phase structure and avoids the phase distortion attributed to Fourier-domain operations.
Section 3.2 states this without proof; if false, the claimed spatial-cue benefit of WTConv is unsupported.
domain assumption Matching the MUSIC spatial spectrum of the enhanced signal to that of the noisy input is a valid proxy for preserving target-speech spatial cues.
Section 4.3 loss Lps does exactly this, while evaluation uses delta ITD/IPD/ILD; the link between the two is assumed, not established.
domain assumption Image-method simulated RIRs adequately represent real-world spatial cues.
Section 4.1 uses simulated rooms only; generalization to measured arrays is untested.
ad hoc to paper Removing the filter-and-sum step from EaBNet produces a valid MIMO comparison system.
Section 4.4 states the removal without describing the resulting output or validating that it is a fair MIMO baseline.
standard math STFT, Haar wavelet, MUSIC, and SI-SNR definitions are standard and correctly applied.
Used throughout Sections 2-4 without derivation.

how reviews work

0 comments

Cite this review

Pith. "Pith review of WTFormer: A Wavelet Conformer Network for MIMO Speech Enhancement with Spatial Cues Peservation." pith.science (2026). https://pith.science/paper/ORYOYK2S

@misc{pith2026250622001,
  author       = {Pith},
  title        = {Pith review of: WTFormer: A Wavelet Conformer Network for MIMO Speech Enhancement with Spatial Cues Peservation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ORYOYK2S}},
  note         = {Machine review of arXiv:2506.22001}
}

read the original abstract

Current multi-channel speech enhancement systems mainly adopt single-output architecture, which face significant challenges in preserving spatio-temporal signal integrity during multiple-input multiple-output (MIMO) processing. To address this limitation, we propose a novel neural network, termed WTFormer, for MIMO speech enhancement that leverages the multi-resolution characteristics of wavelet transform and multi-dimensional collaborative attention to effectively capture globally distributed spatial features, while using Conformer for time-frequency modeling. A multi task loss strategy accompanying MUSIC algorithm is further proposed for optimization training to protect spatial information to the greatest extent. Experimental results on the LibriSpeech dataset show that WTFormer can achieve comparable denoising performance to advanced systems while preserving more spatial information with only 0.98M parameters.

Figures

Figures reproduced from arXiv: 2506.22001 by the authors.

**Figure 1.** An overview of the proposed WTFormer architecture.Different modules are remarked with different colors. with frequency index f ∈ {1, · · · , F} and time index t ∈ {1, · · · , T}, respectively. Hs and Hn denote multichannel relative transfer function (RTF) representing speech and noise. Sf,t and Nf,t represent speech and noise source signals. Using early-reverberation as learning target is better than using directp… view at source ↗

**Figure 2.** [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 31 canonical work pages

[1]

Traditional single-channel methods [1, 2] relied on signal processing and statistical modeling

Introduction Speech enhancement aims to recover clean target speech from noisy mixtures. Traditional single-channel methods [1, 2] relied on signal processing and statistical modeling. However, these algorithms suffer great performance degradation in non station- ary noise and low signal-noise ration (SNR) scenarios. Multi- channel beamforming algorithms ...

work page
[2]

WTFormer: A Wavelet Conformer Network for MIMO Speech Enhancement with Spatial Cues Peservation

Signal Model and Problem Formulation The signal recorded by a uniform linear M-channel micro- phone array can be expressed in the short-time Fourier trans- form (STFT) domain as: Yf,t = Sf,t + Nf,t = HsSf,t + HnNf,t, (1) where{Yf,t, Sf,t, Nf,t} ∈CM denotes the reverberant-noisy mixture speech, target speech and noise for M channels, arXiv:2506.22001v1 [ee...

work page Pith review arXiv 2025
[3]

System overview The model takes multi-channel input in the form of time- frequency domain representations of noisy speech signals

Proposed WTFormer 3.1. System overview The model takes multi-channel input in the form of time- frequency domain representations of noisy speech signals. Ini- tially, the time-domain signal is processed using the short-time Fourier transform (STFT) to extract time-frequency features. We adopt convolutional encoder-decoder (CED) structure with skip connect...

work page
[4]

All constituent sub-blocks incorporate residual connections to maintain gradi- ent flow and preserve original signal fidelity

a final point-wise convolution with dropout. All constituent sub-blocks incorporate residual connections to maintain gradi- ent flow and preserve original signal fidelity. 3.4. MCA Block In this paper, we proposes to use multidimensional collabora- tive attention module (MCA) [24] to replace the traditional skip connection structure. The MCA block utilize...

work page
[5]

Dataset Preparation We used the public speech dataset LibriSpeech and multi- channel RIR to generate microphone-array signals for experi- ments

Experiment 4.1. Dataset Preparation We used the public speech dataset LibriSpeech and multi- channel RIR to generate microphone-array signals for experi- ments. The uniform linear array (ULA) with 4 cm space in- terval and eight elements was used. The train-360 corpus was randomly split: 90% for training, 5% for verification, and 5% for evaluation. The mu...

work page
[6]

Results and Discussion 5.1. Ablation Study We conduct the ablation study on WTFormer as shown in Ta- ble 1, where WTFormer-WT, WTFormer-MCA, WTFormer- Lps indicate the removal of the WTConv, MCA, and Lps loss, respectively. It can be seen that ablation of WTConv slightly degrades the ∆ITD, but greatly affects the PESQ scores. This implies that WTConv impr...

work page
[7]

Conclusions This paper introduces WTFormer, a novel MIMO speech en- hancement framework that preserves spatial cues while achiev- ing competitive noise reduction. By integrating wavelet convo- lutions for multi-resolution analysis, TF-Conformer blocks for time-frequency modeling, and multidimensional collaborative attention for spatial dependency learning...

work page
[8]

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

1979

Show all 39 references

[9]

Nonnegative matrix fac- torization with the Itakura-Saito divergence: With application to music analysis,

C. F ´evotte, N. Bertin, and J.-L. Durrieu, “Nonnegative matrix fac- torization with the Itakura-Saito divergence: With application to music analysis,” Neural computation, vol. 21, no. 3, pp. 793–830, 2009

2009
[10]

High-resolution frequency-wavenumber spectrum analysis,

J. Capon, “High-resolution frequency-wavenumber spectrum analysis,” Proceedings of the IEEE, vol. 57, no. 8, pp. 1408–1418, 1969

1969
[11]

An alternative approach to linearly con- strained adaptive beamforming,

L. Griffiths and C. Jim, “An alternative approach to linearly con- strained adaptive beamforming,” IEEE Transactions on antennas and propagation, vol. 30, no. 1, pp. 27–34, 1982

1982
[12]

Sixty years of frequency-domain monaural speech en- hancement: From traditional to deep learning methods,

C. Zheng, H. Zhang, W. Liu, X. Luo, A. Li, X. Li, and B. C. Moore, “Sixty years of frequency-domain monaural speech en- hancement: From traditional to deep learning methods,”Trends in Hearing, vol. 27, p. 23312165231209913, 2023

2023
[13]

Neural network based spectral mask estimation for acoustic beamforming,

J. Heymann, L. Drude, and R. Haeb-Umbach, “Neural network based spectral mask estimation for acoustic beamforming,” in In- ternational Conference on Acoustics, Speech and Signal Process- ing (ICASSP). IEEE, 2016, pp. 196–200

2016
[14]

Improved MVDR beamforming using single-channel mask prediction networks,

H. Erdogan, J. R. Hershey, S. Watanabe, M. I. Mandel, and J. Le Roux, “Improved MVDR beamforming using single-channel mask prediction networks,” in Interspeech, 2016, pp. 1981–1985

2016
[15]

FaS- Net: Low-latency adaptive beamforming for multi-microphone audio processing,

Y . Luo, C. Han, N. Mesgarani, E. Ceolini, and S.-C. Liu, “FaS- Net: Low-latency adaptive beamforming for multi-microphone audio processing,” in IEEE automatic speech recognition and un- derstanding workshop (ASRU). IEEE, 2019, pp. 260–267

2019
[16]

Neural spectrospatial filter- ing,

K. Tan, Z.-Q. Wang, and D. Wang, “Neural spectrospatial filter- ing,” IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 30, pp. 605–621, 2022

2022
[17]

Phase-aware speech enhancement based on deep neural networks,

N. Zheng and X.-L. Zhang, “Phase-aware speech enhancement based on deep neural networks,” IEEE/ACM Transactions on Au- dio, Speech, and Language Processing, vol. 27, no. 1, pp. 63–76, 2018

2018
[18]

Real-time binaural speech separation with preserved spatial cues,

C. Han, Y . Luo, and N. Mesgarani, “Real-time binaural speech separation with preserved spatial cues,” in International Con- ference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2020, pp. 6404–6408

2020
[19]

MIMO noise suppression preserving spatial cues for sound source localization in mobile robot,

J.-H. Kim, J. Choi, J. Son, G.-S. Kim, J. Park, and J.-H. Chang, “MIMO noise suppression preserving spatial cues for sound source localization in mobile robot,” in International Symposium on Circuits and Systems (ISCAS). IEEE, 2021, pp. 1–5

2021
[20]

Complex-valued spatial au- toencoders for multichannel speech enhancement,

M. M. Halimeh and W. Kellermann, “Complex-valued spatial au- toencoders for multichannel speech enhancement,” in Interna- tional Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2022, pp. 261–265

2022
[21]

Diffusion model-based MIMO speech denoising and dereverberation,

R. Kimura, T. Nakatani, N. Kamo, D. Marc, S. Araki, T. Ueda, and S. Makino, “Diffusion model-based MIMO speech denoising and dereverberation,” in International Conference on Acoustics, Speech, and Signal Processing Workshops (ICASSPW) . IEEE, 2024, pp. 455–459

2024
[22]

A Causal U-Net Based Neural Beamforming Network for Real-Time Multi-Channel Speech Enhancement,

X. Ren, X. Zhang, L. Chen, X. Zheng, C. Zhang, L. Guo, and B. Yu, “A Causal U-Net Based Neural Beamforming Network for Real-Time Multi-Channel Speech Enhancement,” in Interspeech, 2021, pp. 1832–1836

2021
[23]

TPARN: Triple-path attentive recurrent network for time-domain multichannel speech enhancement,

A. Pandey, B. Xu, A. Kumar, J. Donley, P. Calamia, and D. Wang, “TPARN: Triple-path attentive recurrent network for time-domain multichannel speech enhancement,” in International Conference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2022, pp. 6497–6501

2022
[24]

Multi-microphone complex spectral mapping for utterance-wise and continuous speech sepa- ration,

Z.-Q. Wang, P. Wang, and D. Wang, “Multi-microphone complex spectral mapping for utterance-wise and continuous speech sepa- ration,” IEEE/ACM transactions on audio, speech, and language processing, vol. 29, pp. 2001–2014, 2021

2001
[25]

Binaural Speech Enhancement Using Deep Com- plex Convolutional Transformer Networks,

V . Tokala, E. Grinstein, M. Brookes, S. Doclo, J. Jensen, and P. A. Naylor, “Binaural Speech Enhancement Using Deep Com- plex Convolutional Transformer Networks,” inInternational Con- ference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2024, pp. 681–685

2024
[26]

Cmgan: Conformer- based metric-gan for monaural speech enhancement,

S. Abdulatif, R. Cao, and B. Yang, “Cmgan: Conformer- based metric-gan for monaural speech enhancement,” IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2024

2024
[27]

Embedding and beamform- ing: All-neural causal beamformer for multichannel speech en- hancement,

A. Li, W. Liu, C. Zheng, and X. Li, “Embedding and beamform- ing: All-neural causal beamformer for multichannel speech en- hancement,” in International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2022, pp. 6487–6491

2022
[28]

A systematic study of DNN based speech enhancement in reverberant and reverberant- noisy environments,

H. Wang, A. Pandey, and D. Wang, “A systematic study of DNN based speech enhancement in reverberant and reverberant- noisy environments,” Computer Speech & Language , vol. 89, p. 101677, 2025

2025
[29]

Wavelet convolutions for large receptive fields,

S. E. Finder, R. Amoyal, E. Treister, and O. Freifeld, “Wavelet convolutions for large receptive fields,” in European Conference on Computer Vision. Springer, 2024, pp. 363–380

2024
[30]

Conformer: Convolution- augmented transformer for speech recognition,

A. Gulati, J. Qin, C.-C. Chiu, N. Parmar, Y . Zhang, J. Yu, W. Han, S. Wang, Z. Zhang, Y . Wu et al. , “Conformer: Convolution- augmented transformer for speech recognition,” arXiv preprint arXiv:2005.08100, 2020

2005 arXiv
[31]

MCA: Mul- tidimensional collaborative attention in deep convolutional neural networks for image recognition,

Y . Yu, Y . Zhang, Z. Cheng, Z. Song, and C. Tang, “MCA: Mul- tidimensional collaborative attention in deep convolutional neural networks for image recognition,”Engineering Applications of Ar- tificial Intelligence, vol. 126, p. 107079, 2023

2023
[32]

Complex ratio mask- ing for monaural speech separation,

D. S. Williamson, Y . Wang, and D. Wang, “Complex ratio mask- ing for monaural speech separation,” IEEE/ACM transactions on audio, speech, and language processing, vol. 24, no. 3, pp. 483– 492, 2015

2015
[33]

Adam: A method for stochastic optimization,

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

2014 arXiv
[34]

Mixed precision training,

P. Micikevicius, S. Narang, J. Alben, G. Diamos, E. Elsen, D. Gar- cia, B. Ginsburg, M. Houston, O. Kuchaiev, G. Venkatesh et al., “Mixed precision training,” arXiv preprint arXiv:1710.03740 , 2017

2017 arXiv
[35]

Multi-task learning using uncertainty to weigh losses for scene geometry and semantics,

A. Kendall, Y . Gal, and R. Cipolla, “Multi-task learning using uncertainty to weigh losses for scene geometry and semantics,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 7482–7491

2018
[36]

Single-channel multi-speaker separation using deep clustering,

Y . Isik, J. L. Roux, Z. Chen, S. Watanabe, and J. R. Hershey, “Single-channel multi-speaker separation using deep clustering,” arXiv preprint arXiv:1607.02173, 2016

2016 arXiv
[37]

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 international conference on acoustics, speech, and signal process- ing (ICASSP), vol....

2001
[38]

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

J. Jensen and C. H. Taal, “An algorithm for predicting the in- telligibility of speech masked by modulated noise maskers,” IEEE/ACM Transactions on Audio, Speech, and Language Pro- cessing, vol. 24, no. 11, pp. 2009–2022, 2016

2009
[39]

SDR– half-baked or well done?

J. Le Roux, S. Wisdom, H. Erdogan, and J. R. Hershey, “SDR– half-baked or well done?” in International Conference on Acous- tics, Speech and Signal Processing (ICASSP) . IEEE, 2019, pp. 626–630

2019

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

Traditional single-channel methods [1, 2] relied on signal processing and statistical modeling

Introduction Speech enhancement aims to recover clean target speech from noisy mixtures. Traditional single-channel methods [1, 2] relied on signal processing and statistical modeling. However, these algorithms suffer great performance degradation in non station- ary noise and low signal-noise ration (SNR) scenarios. Multi- channel beamforming algorithms ...

work page

[2] [2]

WTFormer: A Wavelet Conformer Network for MIMO Speech Enhancement with Spatial Cues Peservation

Signal Model and Problem Formulation The signal recorded by a uniform linear M-channel micro- phone array can be expressed in the short-time Fourier trans- form (STFT) domain as: Yf,t = Sf,t + Nf,t = HsSf,t + HnNf,t, (1) where{Yf,t, Sf,t, Nf,t} ∈CM denotes the reverberant-noisy mixture speech, target speech and noise for M channels, arXiv:2506.22001v1 [ee...

work page Pith review arXiv 2025

[3] [3]

System overview The model takes multi-channel input in the form of time- frequency domain representations of noisy speech signals

Proposed WTFormer 3.1. System overview The model takes multi-channel input in the form of time- frequency domain representations of noisy speech signals. Ini- tially, the time-domain signal is processed using the short-time Fourier transform (STFT) to extract time-frequency features. We adopt convolutional encoder-decoder (CED) structure with skip connect...

work page

[4] [4]

All constituent sub-blocks incorporate residual connections to maintain gradi- ent flow and preserve original signal fidelity

a final point-wise convolution with dropout. All constituent sub-blocks incorporate residual connections to maintain gradi- ent flow and preserve original signal fidelity. 3.4. MCA Block In this paper, we proposes to use multidimensional collabora- tive attention module (MCA) [24] to replace the traditional skip connection structure. The MCA block utilize...

work page

[5] [5]

Dataset Preparation We used the public speech dataset LibriSpeech and multi- channel RIR to generate microphone-array signals for experi- ments

Experiment 4.1. Dataset Preparation We used the public speech dataset LibriSpeech and multi- channel RIR to generate microphone-array signals for experi- ments. The uniform linear array (ULA) with 4 cm space in- terval and eight elements was used. The train-360 corpus was randomly split: 90% for training, 5% for verification, and 5% for evaluation. The mu...

work page

[6] [6]

Results and Discussion 5.1. Ablation Study We conduct the ablation study on WTFormer as shown in Ta- ble 1, where WTFormer-WT, WTFormer-MCA, WTFormer- Lps indicate the removal of the WTConv, MCA, and Lps loss, respectively. It can be seen that ablation of WTConv slightly degrades the ∆ITD, but greatly affects the PESQ scores. This implies that WTConv impr...

work page

[7] [7]

Conclusions This paper introduces WTFormer, a novel MIMO speech en- hancement framework that preserves spatial cues while achiev- ing competitive noise reduction. By integrating wavelet convo- lutions for multi-resolution analysis, TF-Conformer blocks for time-frequency modeling, and multidimensional collaborative attention for spatial dependency learning...

work page

[8] [8]

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

1979