Lightweight DNN for Full-Band Speech Denoising on Mobile Devices: Exploiting Long and Short Temporal Patterns

Konstantinos Drossos; Mikko Heikkinen; Paschalis Tsiaflakis

REVIEW 4 major objections 5 minor 26 references

Lightweight DNN for Full-Band Speech Denoising on Mobile Devices: Exploiting Long and Short Temporal Patterns

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

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

Pith's one-line read A causal 48 kHz denoiser with 0.45M parameters is reported to beat full-band baselines by over 6 dB SI-SDR while running at real-time factor 0.014 on a phone CPU.

desk verdict Solid engineering result with a credible mobile full-band denoiser, but the headline SOTA claim rests on cross-protocol baseline numbers and an arithmetic slip (5.71 dB, not 'over 6'). read the letter →

arxiv 2509.05079 v1 pith:OCQNA5BK submitted 2025-09-05 eess.AS cs.LGcs.SDeess.SP

Konstantinos Drossos , Mikko Heikkinen , Paschalis Tsiaflakis This is my paper · ORCID

classification eess.AScs.LGcs.SDeess.SP

keywords speechdenoisingfull-bandaudiolowlatencyreal-timeUNetrecurrentneuralnetworksmobiledevicesSI-SDR

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper sets out to show that full-band (48 kHz) speech denoising can be done in real time on a mobile phone without giving up quality. It builds a causal, low-latency DNN that processes one short-time Fourier transform (STFT) magnitude frame at a time, with no look-ahead frames, and reports the highest scale-invariant signal-to-distortion ratio (SI-SDR) among full-band methods it compares against: 22.34 dB on the VCTK test set, more than 6 dB above published results for PercepNet, DeepFilterNet v1, and DeepFilterNet v3. That result comes with 0.45M parameters, 0.0064 GMACs, and a real-time factor of 0.014 on a phone CPU, roughly 100x fewer MACs and about 4x lower latency than the compared baselines. If the comparison holds, this closes the usual gap between high-quality full-band denoising and deployment on resource-constrained devices.

What carries the argument

The load-bearing mechanism is a pair of temporal-context techniques used twice. In the front end, 32 look-back frames of the mapped STFT magnitude are stacked along time and folded into the channel dimension by temporally-spanning convolutional kernels, capturing short patterns; in the bottleneck and again just before mask prediction, single-layer GRUs carry state across frames, capturing long patterns. The resulting causal system never sees a future frame. Complexity is kept low by a learned input mapping that shrinks the full STFT to 96 features, MobileNet-v2-style inverted bottlenecks in the encoder, and mask prediction in that reduced space, with the mask then mapped back to all frequenc

What would settle it

Run the three baselines through the exact test pipeline used here (VCTK test set, SNR sampled uniformly from -10 to 25 dB, up to two noise files mixed, peak amplitude scaled to 0.001-0.999, 2048-point STFT, and the same SI-SDR implementation). If the proposed model's SI-SDR does not exceed the baselines by roughly 6 dB, or falls below them, the central claim is refuted.

Watch

Extended reading notes

Core claim

The central claim is that a deliberately minimal, causal full-band denoiser can exceed the SI-SDR of established full-band methods while being much smaller and faster. The model takes the current STFT magnitude frame plus 32 prior mapped frames, encodes them with a UNet-style stack whose kernels span the look-back window, runs a GRU bottleneck to carry long-term state, decodes with skip connections, then passes the representation through a second small GRU autoencoder before predicting a sigmoid mask in a reduced 96-dimensional space. The mask multiplies the noisy magnitude; phase passes through unchanged. On the VCTK test set the authors report 22.34 dB SI-SDR, 0.94 STOI, and 2.82 PESQ-WB,

Load-bearing premise

The headline 6 dB SI-SDR advantage assumes the PercepNet, DeepFilterNet v1, and DeepFilterNet v3 numbers quoted from other papers came from the same test protocol—same SNR sampling, noise mixing, peak scaling, and SI-SDR code—as the proposed method; if the published baselines were evaluated under different conditions, the reported gap could be an artifact of protocol mismatch rather than model superiority.

Editorial extensions

If this is right

A causal, no-look-ahead model can deliver SI-SDR gains of more than 6 dB over full-band baselines, so real-time voice chains no longer need to sacrifice denoising quality to stay causal.
At 0.45M parameters and 0.0064 GMACs, full-band 48 kHz denoising fits in mobile budgets and can run on a phone CPU at a real-time factor of 0.014, leaving headroom for the rest of the processing chain.
Removing the learned input/output mapping cuts the real-time factor to about 0.012, showing that the mapping itself contributes roughly 14% of the runtime cost.
The reported gap between high SI-SDR and lower PESQ-WB suggests that full-band methods need metrics that reward the 8-24 kHz region, not just narrowband quality.

Reading between the lines

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

If the 6 dB SI-SDR gap survives a shared evaluation protocol, temporal modeling (look-back plus two GRU stages) is likely the main driver of the gain, meaning similar advantages might transfer to other full-band tasks such as dereverberation or acoustic echo cancellation.
A direct test of the metric hypothesis would be to recompute SI-SDR after low-pass filtering both the estimate and reference to 8 kHz; if the gap mostly remains, the PESQ-WB discrepancy is not simply a bandwidth artifact.
Because phase is unmodified and only magnitude is masked, the method can be dropped into an existing codec or enhancement pipeline; measuring end-to-end quality, rather than isolated denoising quality, would show whether the low perceptual scores matter in practice.
The learned input mapping to 96 features buys speech fidelity at a roughly 14% speed cost; replacing it with a fixed filterbank in an ablation would separate the benefit of learned compression from the benefit of the temporal architecture.

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.

Desk Editor's Note

read the letter

The takeaway: this is a useful engineering paper, not a scientific breakthrough. The authors build a 0.45M-parameter causal full-band (48 kHz) denoiser that runs on a phone with RTF 0.014 and reports 22.34 dB SI-SDR on VCTK. If those numbers survive reproduction, the architecture is a practical addition to the mobile voice stack. The design is a sensible combination of known pieces—UNet with MobileNet inverted bottlenecks, look-back frames, a GRU bottleneck, and a GRU-based mask autoencoder. That combination is new in this exact form, and the paper describes it clearly enough to reimplement.

The evaluation is mostly sound: public datasets, an 80/20 split, early stopping, and the authors are careful to exclude VCTK overlap between training and test. The DNSMOS-based filtering of training data is a reasonable pre-processing choice. They also explicitly note that the RTF comparison across devices might be pessimistic for their method, which is honest.

The soft spots are real but fixable. The headline claim—'highest SI-SDR, over 6 dB'—rests on comparing against DFv1's published value of 16.63, which was likely computed under a different test protocol (different SNR distribution, peak scaling, or evaluation code). The paper does not rerun the baselines, reports no variance, and does no significance testing. The arithmetic is also off: 22.34 - 16.63 = 5.71 dB, not over 6 dB. That matters because the text makes the claim explicitly. Also, their PESQ-WB and DNSMOS are lower than DeepFilterNet v3; they conjecture this is because those metrics are 16 kHz-centric, which is plausible but not demonstrated. That's an honest limitation, not a deception.

Who is this for? Practitioners in real-time full-band enhancement on constrained hardware. A serious referee should handle it; the core feasibility claim is credible, and the engineering detail is worth scrutiny. The authors should be asked to recompute or clearly state the baseline protocol, report standard deviations, fix the arithmetic, and ideally release a checkpoint.

Referee Report

4 major / 5 minor

Summary. The paper proposes a causal, full-band (48 kHz) speech denoising DNN intended for mobile deployment. The architecture is a U-Net variant with a learned input/output mapping, look-back frames at the encoder and mask predictor, temporally spanning convolutional kernels, and GRU-based bottlenecks, yielding 0.451M parameters, 0.0064 GMACs/s, and a measured TFLite RTF of 0.014 on a Pixel 7. Training uses SI-SDR as the loss on DNS-Challenge data filtered by DNSMOS, with VCTK as the test set. The paper reports SI-SDR of 22.34 dB, STOI 0.94, PESQ-WB 2.82, and claims a >6 dB SI-SDR advantage over DeepFilterNet v1 while being far lighter and faster. The authors also conjecture why PESQ-WB and DNSMOS are lower than DeepFilterNet v3.

Significance. If the headline claim survives scrutiny, the paper would be a genuinely useful engineering contribution: it demonstrates that a very small causal model can achieve SI-SDR competitive with or exceeding much larger full-band systems at 48 kHz, and it backs the deployability claim with an actual TFLite benchmark on a mobile phone. The explicit design choices--look-back frames for short temporal context, GRUs for long context, MobileNet-style bottlenecks, causal instance normalization--are clearly described and mostly reproducible from the text. The paper also uses public datasets and a conventional train/validation/test split with early stopping, which is good empirical hygiene. The main weakness is the comparison protocol: the baseline numbers are quoted from other papers rather than recomputed under the same test conditions, and the paper gives no variance or significance information for its own single-run results. This makes the central 'over 6 dB' claim fragile regardless of how carefully the proposed model is evaluated.

major comments (4)

[Section III-C and Table I] The SI-SDR baseline for DeepFilterNet v1 (16.63 dB) is taken from the DeepFilterNet paper, not recomputed under the protocol used for the proposed method. The training/test protocol in Section III-A differs from other works in SNR range (-10 to 25 dB), peak scaling (0.001-0.999), noise mixing (up to two noise files), and STFT parameters. Since SI-SDR is sensitive to these choices, the 5.71 dB gap shown in Table I could be partly a protocol artifact. The authors should either recompute all baselines under the identical evaluation script/protocol or clearly label the comparison as cross-paper and temper the claim accordingly.
[Section IV, 'over 6 dB' claim] The text states 'our method provides the highest SI-SDR with a difference of over 6 dB.' Table I gives Ours = 22.34 and DFv1 = 16.63, so the difference is 5.71 dB, not over 6 dB. This arithmetic discrepancy matters because the >6 dB gap is the headline quantitative claim. It should be corrected and the claim restated based on the actual recomputed or cross-paper numbers.
[Section III-C and Table I (single-run, no uncertainty)] The paper reports a single run for the proposed method and no confidence intervals, error bars, or significance tests for any metric. Given that the central claim is a numerical advantage over baselines, the absence of any measure of variability (e.g., multiple training runs, bootstrap intervals, or paired significance tests) weakens the support. At minimum, the authors should state whether the result is from one run, report the validation performance, and add variance estimates or statistical testing.
[Section III-C and Table II] The RTF comparison is across different hardware: Ours is measured on a Pixel 7 phone, while PercepNet and DFv1 values are taken from a laptop-based reference. The Table II note acknowledges this, but the conclusion 'almost as four times faster' is not supported by a like-for-like comparison. The authors should either benchmark the baselines on the same mobile device or explicitly delimit the claim to 'RTF on different hardware' and avoid direct speed superiority statements.

minor comments (5)

[Section IV] The explanation for lower PESQ-WB (that PESQ only evaluates up to 8 kHz and penalizes fullband artifacts) is presented as a likely cause but is speculative. If the authors want to make this claim, they should provide supporting evidence, e.g., band-limited analysis or an ablation.
[Section III-A] The DNSMOS filtering criterion is given, but the DNSMOS model version and the exact threshold rationale are not discussed. Also, the paper does not state how many files were removed by filtering.
[Section III-B] The phrase 'having 16 Hz sampling rate' in the Introduction appears to be a typo for 16 kHz. Similarly, 'denoinsing' and 'wtih' typos appear in the text.
[Section III-C] The comparison with PercepNet is limited because PercepNet is not evaluated on SI-SDR or STOI in Table I; only PESQ-WB is listed. This makes the 'surpasses all but one' claim only partially supported.
[Section II] The notation for the dimensions of H_in-pad would benefit from a concrete tensor-shape example. Also, the exact implementation of the 'causal instance norm' (whether it is a running-statistics or batch-dependent normalization) is not fully specified.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the training/evaluation split is clean and the headline comparison rests on external baselines, not on fitted inputs or self-citation.

full rationale

This is standard supervised empirical work, not a derivation. The model is trained to minimize SI-SDR between the synthesized enhanced signal and the clean reference, and SI-SDR is also used as an evaluation metric; this is normal loss/evaluation alignment, not circularity, because the VCTK test set is external and was excluded from training and validation (the paper explicitly states overlapping VCTK files were removed). The only mild overlap is that DNSMOS is used to filter the dev-set and also appears as an evaluation metric, but the test set is not DNSMOS-filtered and the DNSMOS scores are measured on held-out test utterances, so they are not forced by construction. The comparison values for PercepNet, DeepFilterNet v1, and v3 are quoted from other papers and are not recomputed under the same protocol; while this is a real threat to the validity of the 'over 6 dB' claim and a correctness/experimental-design risk, it is not circular reasoning in the paper's own derivation. There are no load-bearing self-citations, no imported uniqueness theorems, and no ansatz smuggled in via citation. Therefore the circularity score is 0.

Assumptions & free parameters 9 free parameters · 4 assumptions · 0 invented entities

The central claim rests on standard supervised training assumptions: the chosen architecture, hyperparameters, loss, and data filtering are asserted to produce a useful denoiser. The ledger lists the tuned hyperparameters and domain assumptions that are not independently justified in the paper.

free parameters (9)

F'=96 mapping size = 96
Learned input/output mapping reduces frequency dimension; value chosen by validation.
T_pad=32 look-back frames for encoder = 32
Number of previous mapped input frames fed to encoder; chosen by validation.
T_M=3 look-back frames for mask predictor = 3
Number of previous frames fed to mask predictor; chosen by validation.
N_E=N_D=6 encoder/decoder blocks = 6
Depth of UNet backbone; chosen by validation.
Encoder channels C=256, C_out=16 (final 64) = 256/16/64
MobileNet-style bottleneck widths; chosen by validation.
Kernel sizes and strides [5,3,5,3,5,3] / [2,1,2,1,2,2] = listed
Temporal spans and downsampling schedule; chosen by validation.
STFT size 2048, 50% overlap = 2048/1024 hop
Time-frequency resolution; chosen by hand.
SNR range -10 to 25 dB and peak scaling 0.001 to 0.999 = listed
Training data mixing distribution; chosen by hand.
Training hyperparameters (batch 32, lr 1e-4, grad clip 0.5, patience 200) = listed
Optimization choices; chosen by monitoring validation.

assumptions (4)

domain assumption Phase of the noisy signal can be reused unaltered.
Equation (11) applies a mask to magnitude while the noisy phase synthesizes the output; if phase errors dominate at 48 kHz, the method cannot correct them. Section II-E.
domain assumption DNS-Challenge training distribution transfers to the VCTK test set.
Training/validation on DNS-Challenge, test on VCTK; no domain adaptation or calibration. Section III-A.
domain assumption DNSMOS thresholds (MOS-SIG >= 4.0, MOS-OVRL >= 3.9) remove only low-quality files and do not distort the training distribution.
Section III-A filters the dev-set; if filtering selects easy examples, reported test scores may not reflect typical conditions.
domain assumption SI-SDR is a suitable training loss and evaluation metric for full-band speech.
Used as loss and headline metric; the paper acknowledges PESQ-WB and DNSMOS disagree with SI-SDR ordering. Section IV.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Lightweight DNN for Full-Band Speech Denoising on Mobile Devices: Exploiting Long and Short Temporal Patterns." pith.science (2026). https://pith.science/paper/OCQNA5BK

@misc{pith2026250905079,
  author       = {Pith},
  title        = {Pith review of: Lightweight DNN for Full-Band Speech Denoising on Mobile Devices: Exploiting Long and Short Temporal Patterns},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OCQNA5BK}},
  note         = {Machine review of arXiv:2509.05079}
}

read the original abstract

Speech denoising (SD) is an important task of many, if not all, modern signal processing chains used in devices and for everyday-life applications. While there are many published and powerful deep neural network (DNN)-based methods for SD, few are optimized for resource-constrained platforms such as mobile devices. Additionally, most DNN-based methods for SD are not focusing on full-band (FB) signals, i.e. having 48 kHz sampling rate, and/or low latency cases. In this paper we present a causal, low latency, and lightweight DNN-based method for full-band SD, leveraging both short and long temporal patterns. The method is based on a modified UNet architecture employing look-back frames, temporal spanning of convolutional kernels, and recurrent neural networks for exploiting short and long temporal patterns in the signal and estimated denoising mask. The DNN operates on a causal frame-by-frame basis taking as an input the STFT magnitude, utilizes inverted bottlenecks inspired by MobileNet, employs causal instance normalization for channel-wise normalization, and achieves a real-time factor below 0.02 when deployed on a modern mobile phone. The proposed method is evaluated using established speech denoising metrics and publicly available datasets, demonstrating its effectiveness in achieving an (SI-)SDR value that outperforms existing FB and low latency SD methods.

Figures

Figures reproduced from arXiv: 2509.05079 by the authors.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 22 canonical work pages

[1]

Segan: Speech enhancement generative adversarial network,

S. Pascual, A. Bonafonte, and J. Serr `a, “Segan: Speech enhancement generative adversarial network,” 2017. [Online]. Available: https: //arxiv.org/abs/1703.09452

arXiv 2017
[2]

Deepfil- ternet: A low complexity speech enhancement framework for full-band audio based on deep filtering,

H. Schroter, A. N. Escalante-B, T. Rosenkranz, and A. Maier, “Deepfil- ternet: A low complexity speech enhancement framework for full-band audio based on deep filtering,” inIEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2022

work page 2022
[3]

Conv-TasNet: Surpassing Ideal Time–Frequency Magnitude Masking for Speech Separation,

Y . Luo and N. Mesgarani, “Conv-TasNet: Surpassing Ideal Time–Frequency Magnitude Masking for Speech Separation,” IEEE/ACM Trans. Audio, Speech and Lang. Proc., vol. 27, no. 8, p. 1256–1266, Aug. 2019

work page 2019
[4]

Ks-net: Multi-band joint speech restoration and enhance- ment network for 2024 icassp ssi challenge,

G. Yuet al., “Ks-net: Multi-band joint speech restoration and enhance- ment network for 2024 icassp ssi challenge,” inIEEE International Conference on Acoustics, Speech, and Signal Processing Workshops (ICASSPW), 2024

work page 2024
[5]

Immersive voice and audio services (ivas) codec-the new 3gpp standard for immersive communication,

M. Multrus, S. Bruhn, J. Torres, E. Fotopoulou, T. Toftg ˚ard, E. Norvell, S. D ¨ohla, Y . Gao, H.-y. Su, L. Laaksonenet al., “Immersive voice and audio services (ivas) codec-the new 3gpp standard for immersive communication,” in157th AES Convention, 2024

work page 2024
[6]

Real- time denoising and dereverberation wtih tiny recurrent u-net,

H.-S. Choi, S. Park, J. H. Lee, H. Heo, D. Jeon, and K. Lee, “Real- time denoising and dereverberation wtih tiny recurrent u-net,” inIEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2021

work page 2021
[7]

Ultra low complexity deep learning based noise suppression,

S. S. Shetu, S. Chakrabarty, O. Thiergart, and E. Mabande, “Ultra low complexity deep learning based noise suppression,” inIEEE In- ternational Conference on Acoustics, Speech and Signal Processing (ICASSP), 2024

work page 2024
[8]

Gtcrn: A speech enhancement model requiring ultralow computational resources,

X. Rong, T. Sun, X. Zhang, Y . Hu, C. Zhu, and J. Lu, “Gtcrn: A speech enhancement model requiring ultralow computational resources,” inIEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2024

work page 2024

Show all 26 references

[9]

U-Net: Convolutional Net- works for Biomedical Image Segmentation,

O. Ronneberger, P. Fischer, and T. Brox, “U-Net: Convolutional Net- works for Biomedical Image Segmentation,” inMedical Image Comput- ing and Computer-Assisted Intervention (MICCAI 2015), Nov. 2015, pp. 234–241

2015
[10]

A hybrid dsp/deep learning approach to real-time full-band speech enhancement,

J.-M. Valin, “A hybrid dsp/deep learning approach to real-time full-band speech enhancement,”20th IEEE International Workshop on Multimedia Signal Processing (MMSP), 2017

2017
[11]

A perceptually-motivated approach for low-complexity, real-time enhancement of fullband speech,

J.-M. Valin, U. Isik, N. Phansalkar, R. Giri, K. Helwani, and A. Kr- ishnaswamy, “A perceptually-motivated approach for low-complexity, real-time enhancement of fullband speech,” inInterspeech 2020, 2020

2020
[12]

MobileNetV2: Inverted Residuals and Linear Bot- tlenecks,

M. Sandleret al., “MobileNetV2: Inverted Residuals and Linear Bot- tlenecks,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Jun. 2018, pp. 4510–4520

2018
[13]

ICASSP 2023 Deep Noise Suppression Challenge,

H. Dubey, A. Aazami, V . Gopal, B. Naderi, S. Braun, R. Cutler, A. Ju, M. Zohourian, M. Tang, M. Golestaneh, and R. Aichner, “ICASSP 2023 Deep Noise Suppression Challenge,” inIEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2023

2023
[14]

CSTR VCTK Corpus: English Multi-speaker Corpus for CSTR V oice Cloning Toolkit,

C. Veaux, J. Yamagishi, and K. MacDonald, “CSTR VCTK Corpus: English Multi-speaker Corpus for CSTR V oice Cloning Toolkit,” in University of Edinburgh. The Centre for Speech Technology Research (CSTR), 2019

2019
[15]

Investi- gating rnn-based speech enhancement methods for noise-robust text-to- speech,

C. Valentini-Botinhao, X. Wang, S. Takaki, and J. Yamagishi, “Investi- gating rnn-based speech enhancement methods for noise-robust text-to- speech,” in9th ISCA Workshop on Speech Synthesis Workshop (SSW 9), 2016

2016
[16]

DNSMOS: A Non-Intrusive Perceptual Objective Speech Quality metric to evaluate Noise Suppres- sors,

C. K. Reddy, V . Gopal, and R. Cutler, “DNSMOS: A Non-Intrusive Perceptual Objective Speech Quality metric to evaluate Noise Suppres- sors,” inIEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2021

2021
[17]

Searching for MobileNetV3,

A. Howardet al., “Searching for MobileNetV3,” inIEEE/CVF Interna- tional Conference on Computer Vision (ICCV), 2019

2019
[18]

On the properties of neural machine translation: Encoder–decoder approaches,

K. Cho, B. van Merri ¨enboer, D. Bahdanau, and Y . Bengio, “On the properties of neural machine translation: Encoder–decoder approaches,” inProceedings of SSST-8, Eighth Workshop on Syntax, Semantics and Structure in Statistical Translation, D. Wu, M. Carpuat, X. Carreras, and...

2014
[19]

Adam: A method for stochastic optimization,

D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,”
[20]

SDR – Half- baked or Well Done?

J. L. Roux, S. Wisdom, H. Erdogan, and J. R. Hershey, “SDR – Half- baked or Well Done?” inIEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2019

2019
[21]

A short- time objective intelligibility measure for time-frequency weighted noisy speech,

C. H. Taal, R. C. Hendriks, R. Heusdens, and J. Jensen, “A short- time objective intelligibility measure for time-frequency weighted noisy speech,” inIEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2010

2010
[22]

Perceptual evaluation of speech quality (PESQ)-a new method for speech quality assessment of telephone networks and codecs,

A. Rix, J. Beerends, M. Hollier, and A. Hekstra, “Perceptual evaluation of speech quality (PESQ)-a new method for speech quality assessment of telephone networks and codecs,” inIEEE International Conference on Acoustics, Speech, and Signal Processing. Proceedings (ICASSP), 2001

2001
[23]

Deepfilternet: Perceptually motivated real-time speech enhancement,

H. Schr ¨oter, T. Rosenkranz, A. N. Escalante-B., and A. Maier, “Deepfilternet: Perceptually motivated real-time speech enhancement,”
[24]

Deepfil- ternet2: Towards Real-Time Speech Enhancement on Embedded Devices for Full-Band Audio,

H. Schr ¨oter, A. N. Escalante, T. Rosenkranz, and A. K. Maier, “Deepfil- ternet2: Towards Real-Time Speech Enhancement on Embedded Devices for Full-Band Audio,”2022 International Workshop on Acoustic Signal Enhancement (IWAENC), 2022

2022
[2017]

Available: https://arxiv.org/abs/1412.6980

[Online]. Available: https://arxiv.org/abs/1412.6980

arXiv
[2023]

Available: https://arxiv.org/abs/2305.08227

[Online]. Available: https://arxiv.org/abs/2305.08227

arXiv

Pith tools

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

[1] [1]

Segan: Speech enhancement generative adversarial network,

S. Pascual, A. Bonafonte, and J. Serr `a, “Segan: Speech enhancement generative adversarial network,” 2017. [Online]. Available: https: //arxiv.org/abs/1703.09452

arXiv 2017

[2] [2]

Deepfil- ternet: A low complexity speech enhancement framework for full-band audio based on deep filtering,

H. Schroter, A. N. Escalante-B, T. Rosenkranz, and A. Maier, “Deepfil- ternet: A low complexity speech enhancement framework for full-band audio based on deep filtering,” inIEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2022

work page 2022

[3] [3]

Conv-TasNet: Surpassing Ideal Time–Frequency Magnitude Masking for Speech Separation,

Y . Luo and N. Mesgarani, “Conv-TasNet: Surpassing Ideal Time–Frequency Magnitude Masking for Speech Separation,” IEEE/ACM Trans. Audio, Speech and Lang. Proc., vol. 27, no. 8, p. 1256–1266, Aug. 2019

work page 2019

[4] [4]

Ks-net: Multi-band joint speech restoration and enhance- ment network for 2024 icassp ssi challenge,

G. Yuet al., “Ks-net: Multi-band joint speech restoration and enhance- ment network for 2024 icassp ssi challenge,” inIEEE International Conference on Acoustics, Speech, and Signal Processing Workshops (ICASSPW), 2024

work page 2024

[5] [5]

Immersive voice and audio services (ivas) codec-the new 3gpp standard for immersive communication,

M. Multrus, S. Bruhn, J. Torres, E. Fotopoulou, T. Toftg ˚ard, E. Norvell, S. D ¨ohla, Y . Gao, H.-y. Su, L. Laaksonenet al., “Immersive voice and audio services (ivas) codec-the new 3gpp standard for immersive communication,” in157th AES Convention, 2024

work page 2024

[6] [6]

Real- time denoising and dereverberation wtih tiny recurrent u-net,

H.-S. Choi, S. Park, J. H. Lee, H. Heo, D. Jeon, and K. Lee, “Real- time denoising and dereverberation wtih tiny recurrent u-net,” inIEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2021

work page 2021

[7] [7]

Ultra low complexity deep learning based noise suppression,

S. S. Shetu, S. Chakrabarty, O. Thiergart, and E. Mabande, “Ultra low complexity deep learning based noise suppression,” inIEEE In- ternational Conference on Acoustics, Speech and Signal Processing (ICASSP), 2024

work page 2024

[8] [8]

Gtcrn: A speech enhancement model requiring ultralow computational resources,

X. Rong, T. Sun, X. Zhang, Y . Hu, C. Zhu, and J. Lu, “Gtcrn: A speech enhancement model requiring ultralow computational resources,” inIEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2024

work page 2024