FedMLAC: Mutual Learning Driven Heterogeneous Federated Audio Classification

Carlos Busso; Di Wu; Ji Zhang; Jun Bai; Rajib Rana; Shivakumara Palaiahnakote; Xiaohui Tao; Youyang Qu

REVIEW 5 major objections 5 minor 44 references

FedMLAC: Mutual Learning Driven Heterogeneous Federated Audio Classification

T0 review · 5 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Federated audio model beats baselines under noise, skew, and poison

desk verdict Solid empirical integration of mutual learning and robust aggregation for federated audio, but the convergence theorem is not established and key hyperparameters are undisclosed. read the letter →

arxiv 2506.10207 v2 pith:64KLIKCW submitted 2025-06-11 cs.SD cs.DCeess.AS

Jun Bai , Rajib Rana , Di Wu , Youyang Qu , Xiaohui Tao , Ji Zhang , Carlos Busso , Shivakumara Palaiahnakote This is my paper · ORCID

classification cs.SDcs.DCeess.AS

keywords FederatedLearningAudioClassificationMutualKnowledgeDistillationLayer-wisePruningAggregationDataHeterogeneityModelPoisoning

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 one federated learning framework can handle the three problems that jointly degrade federated audio classification: non-identical data distributions across clients, differing client model architectures, and corrupted or poisoned training labels. It proposes FedMLAC, in which every client trains a personal audio model alongside a small shared Plug-in model, and the two teach each other through bidirectional knowledge distillation. The server then aggregates only the Plug-in updates, using a layer-wise pruning step that discards the most extreme parameter deviations. If the claims hold, FedMLAC gives a single recipe that improves accuracy and robustness on four audio benchmarks while preserving client privacy and architectural flexibility.

What carries the argument

The load-bearing object is the lightweight, globally shared Plug-in model, which decouples personalization from aggregation. It is trained on each client together with the client's private model using mutual learning: the local model's task loss is augmented with a KL-divergence term that aligns it with the Plug-in model's predictions, and the Plug-in model in turn is updated with a KL loss that aligns it with the local model, so knowledge flows both ways. The second mechanism is Layer-wise Pruning Aggregation (LPA), which sorts each layer's $\ell^2$ deviation from the cohort average and prunes the largest $\lfloor v_h |\mathcal{S}_t|\rfloor$ and smallest $\lfloor v_l |\mathcal{S}_t|\rfloor$ clients before weighted averaging. LPA's effectiveness enters the convergence proof as a bounded-deviation assumption, and the theorem states that the global Plug-in model reaches a stationary point at rate $\mathcal{O}(1/\sqrt{T})$.

What would settle it

Poison one client's Plug-in model in a single layer while keeping the others clean, then run LPA with the paper's fixed pruning fractions; if the poisoned layer's deviation sits inside the middle subset and the global Plug-in model degrades as much as under plain averaging, Assumption 5 fails in practice. A simpler check is to sweep $v_h$ and $v_l$ from 0 to 30 percent on clean non-IID data: if accuracy peaks sharply only at the hand-picked values and collapses elsewhere, the reported robustness is tuned rather than structural.

Watch

Extended reading notes

Core claim

FedMLAC claims that data heterogeneity, model heterogeneity, and data poisoning can be addressed simultaneously rather than piecemeal. The mechanism is a decoupled design: each client keeps a personalized local audio classification model for its own data and a lightweight global Plug-in model; bidirectional KL-divergence distillation transfers global knowledge to the client model and client knowledge to the Plug-in model. On the server, Layer-wise Pruning Aggregation (LPA) computes each layer's $\ell^2$ deviation from the cohort average, removes the top and bottom fraction of deviating updates, and averages the trusted remainder, so outlier updates are filtered before they enter the global model. The paper reports consistent accuracy gains over FedAvg, FedProx, FedOPT, and FedKAD in both homogeneous and heterogeneous model settings, smaller degradation under additive noise and label errors, and an $\mathcal{O}(1/\sqrt{T})$ convergence bound for the global Plug-in model.

Load-bearing premise

The convergence proof rests on Assumption 5, which says that after Layer-wise Pruning Aggregation prunes the extreme layers, the trusted clients' parameters stay within a bounded deviation $\delta^2$ of the cohort average; if pruning does not actually keep deviations bounded in practice, the theorem's error term and part of the robustness story do not follow.

Editorial extensions

If this is right

FedMLAC is reported to outperform FedAvg, FedProx, FedOPT, and FedKAD on GSC, IEMOCAP, CREMA-D, and Urban Sound, in both homogeneous and heterogeneous model settings.
Under white Gaussian noise down to 10 dB SNR and label error rates up to 0.5, FedMLAC degrades less than every baseline and less than its own LPA-free variant, indicating that LPA carries the robustness gain.
Because clients upload only the lightweight Plug-in model, the framework supports heterogeneous client architectures and partial participation without needing a shared public dataset for distillation.
Theorem 1 states that with learning rate proportional to $1/\sqrt{T}$, the global Plug-in model's average squared gradient norm converges to a stationary point at rate $\mathcal{O}(1/\sqrt{T})$, with the LPA deviation bound $\delta^2$ appearing only in the error term.

Reading between the lines

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

Editorial inference: if per-layer trimmed averaging works as claimed, the same operation could harden other federated aggregation schemes in vision or language, wherever server-side averaging is the bottleneck.
Editorial inference: the paper fixes the pruning fractions $v_h$ and $v_l$; an adaptive per-layer or per-round choice would be a testable extension, since the optimal trusted fraction likely depends on how many clients are actually poisoned.
Editorial inference: because only the Plug-in model is exchanged, the framework may cut communication cost relative to full-model averaging, but the paper does not quantify bytes per round, so that saving is a measurable follow-up rather than an established result.

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

Read it for the empirical story, not the theory. The paper integrates two known ideas—deep mutual learning and layer-wise outlier pruning—into a single federated audio classification framework. Clients train a personalized model and a shared plug-in model with bidirectional KL distillation; the server prunes extreme per-layer deviations before averaging. That specific combination for audio isn't in the prior work I know, and the experiments are broad: four benchmarks, homogeneous and heterogeneous model settings, varying participation, added noise, and label errors. The reported gains are modest (1–3 points) and on IEMOCAP they're inside the standard deviation, so the abstract's "consistently outperforms" overstates things. Still, the ablation with LPA disabled shows the pruning component contributes something real, especially under label noise.

The soft spots are the theory and some practical gaps. Theorem 1 claims an O(1/√T) convergence rate for the global plug-in model, but it doesn't hold as written. Assumption 5 just postulates that LPA keeps trusted-layer deviations bounded by δ², which is basically the result the theorem is supposed to prove. More importantly, the proof jumps from bounds on per-client local plug-in losses to a bound on the gradient of a global objective L(θ^g), but that objective is never defined—Eq. 9 is just a KL loss per client, and aggregation doesn't define an optimization objective for the averaged θ^g. Without a comparison step connecting local gradients to the global gradient at the average, the theorem is not derived from the algorithm. The empirical robustness results are independent of this theorem and stand on their own, so this isn't a load-bearing flaw for the experimental claims, but the theory section shouldn't be cited as proof of convergence.

Other issues: α, v_h, and v_l are never reported, which hurts reproducibility. The FedKAD baseline is modified (public dataset removed), which is defensible but worth flagging. The authors note only the non-convexity and fixed LPA parameters as limitations; the missing objective definition is the bigger gap.

Who's it for: researchers working on personalized or robust federated audio will want this as a baseline. The method is plausible and the experiments are useful. But the theory needs a major rework: either drop the convergence claim or actually define the global plug-in objective and prove a comparison lemma.

My recommendation: accept for peer review—it's a legitimate integration with real experiments and the issues are fixable—but require the hyperparameters and a rewritten theory section. I'd cite it if I were comparing FL audio methods.

Referee Report

5 major / 5 minor

Summary. The paper proposes FedMLAC, a federated learning framework for audio classification that aims to jointly address data heterogeneity, model heterogeneity, and data poisoning. Each client trains a personalized local model together with a lightweight, globally shared Plug-in model, using bidirectional KL-divergence-based mutual learning. The server aggregates Plug-in updates with a Layer-wise Pruning Aggregation (LPA) that removes per-layer outliers before weighted averaging. The authors claim an O(1/sqrt(T)) convergence rate for the global Plug-in model and report experiments on GSC, IEMOCAP, CREMA-D, and Urban Sound under homogeneous and heterogeneous model settings, with robustness tests under additive noise and label errors.

Significance. If the framework works as described, it would be a useful unified contribution to federated audio classification, since most prior work addresses only one or two of the three challenges. The empirical study is fairly broad: four benchmarks, multiple heterogeneity levels, homogeneous and heterogeneous client models, active-ratio sweeps, noise and label-error robustness tests, and an ablation study. The inclusion of an ablated variant without LPA and the consistency of the main accuracy trends are strengths. However, the theoretical convergence result is not established as a theorem about the actual algorithm: it assumes the key property that LPA is supposed to provide, and it does not connect the per-client local loss bounds to the stated global gradient bound. The experimental claims of 'consistently outperforming' baselines are also weakened by missing hyperparameter values and by several comparisons that are within one standard deviation. The core empirical direction is promising, but the paper currently overclaims both theoretically and empirically.

major comments (5)

[§4.1, Assumption 5 and §4.3, Theorem 1 (Eq. 28)] Assumption 5 postulates precisely the property that LPA is supposed to provide: E[||theta^{k,l}_{t+1} - bar_theta^l_{t+1}||^2] <= delta^2 for k in T_l. The pruning operation in Eq. (12) removes extreme deviations from an average that may itself be corrupted; it does not, by itself, bound the deviation of the remaining elements from that average. Since Eq. (28) inherits delta^2 as an error term, the theorem does not independently establish that LPA improves convergence; at best it states convergence under an unverified premise. Please either prove a bound on the trimmed-mean deviation under an explicit corruption model, or clearly state the result as conditional and provide an independent empirical check of the magnitude of delta^2.
[§4.2, Eqs. (27)-(28)] The proof jumps from the per-client local-loss decrease bound in Eq. (27) to the global gradient bound in Eq. (28) without a comparison step. The global objective L(theta^g) minimized by the Plug-in update is never defined: Eq. (9) contains only a KL term and no task loss, so it is unclear what stationary point the bound refers to. A valid derivation needs either a virtual global sequence argument or an explicit bound on ||nabla L(theta^g_t)|| in terms of the trusted local gradients. Without one of these, the O(1/sqrt(T)) rate is not a theorem about FedMLAC as specified in Algorithms 1 and 2.
[§5.2, Hyperparameters and §5.3] The paper never reports the values of alpha, v_h, and v_l, which control the mutual-learning trade-off and the LPA pruning fraction. These are central free parameters of the method, and without them the experiments are not reproducible and the comparison across datasets cannot be checked for inadvertent per-dataset tuning. Please report the exact values used for every experiment, state whether they were tuned per dataset, and include a sensitivity analysis for at least alpha and the pruning rates.
[§5.3, Table 4] The claim that FedMLAC 'consistently outperforms' all baselines is not supported by the reported error bars on several datasets. In Table 4, FedMLAC on IEMOCAP is 51.97 +/- 5.81 versus 50.48 +/- 5.35 for FedKAD and 48.54 +/- 6.59 for FedAvg; these differences are well within one standard deviation of the runs. The same issue appears in several CREMA-D and Urban Sound rows. Please report paired significance tests or confidence intervals across the five runs, or soften the claim to directional improvement without statistical significance.
[§5.2, Baselines (FedKAD modification)] FedKAD is modified by removing its reliance on public datasets and disabling global-logit distillation, leaving only class-wise feature-map communication. This removes a major component of the original FedKAD method and may disadvantage it, making the reported margins over FedKAD potentially inflated. Please justify this modification with an ablation, or compare against the original FedKAD configuration where feasible, and discuss the effect of the modification on the comparison.

minor comments (5)

[§3.1] The section heading 'Prelimineries' is a typo and should read 'Preliminaries'.
[§5.4 and §5.8, Figures 2 and 3] The heterogeneous-model and ablation results are reported only as bar charts, with no numerical table. The text quotes specific improvements (e.g., 'up to 5.23%', 'up to 4.45%'), but readers cannot verify these values or the error bars. Please provide the exact mean +/- standard deviation for every method and configuration.
[§3.1, Eq. (1)] Eq. (1) writes the global objective as a sum over L_k(omega_g), but the local loss is later defined as L_k(omega_k); the argument notation is inconsistent and should be aligned.
[§5.1, Datasets] The 'Urban Sound' dataset is referred to informally; the original dataset name 'UrbanSound8K' should be used in the text and citation for clarity and traceability.
[§5.2, Baselines] The sentence 'For other FedAC methods listed in Table 1, fair comparison is not feasible' is vague; please specify which methods were considered and why they could not be included, or remove the sentence if no attempt was made.

Circularity Check

1 steps flagged · score 3.0 of 10

LPA's convergence benefit is assumed in Assumption 5 and then restated as a conclusion; empirical robustness claims remain independent.

self definitional [Section 4.1, Assumption 5 (Eq. 20); Section 4.4 Discussion]
"Assumption 5 (LPA Effectiveness). The LPA strategy ensures that the parameters of the trusted subset Tl for each layer l have bounded deviation from the average: E[∥θ^{k,l}_{t+1} − θ̄^l_{t+1}∥^2] ≤ δ^2, k ∈ Tl. ... LPA’s pruning strategy bounds δ2 by excluding anomalous updates, enhancing robustness against data poisoning."

The Discussion asserts that LPA bounds δ², but the only support offered is Assumption 5, which simply postulates that bound as 'LPA Effectiveness.' The pruning rule in Eq. (12)–(14) is never shown to imply the bound; the bound is the assumption. The convergence bound in Theorem 1 then carries this assumed δ² directly into Eq. (28), so the theoretical robustness benefit of LPA reduces to the assumption that LPA is effective. The paper is transparent that this is an assumption, but the Discussion presents it as an established property, making this step circular rather than derived.

full rationale

The central empirical claims are not circular: FedMLAC is evaluated against external baselines (FedAvg, FedProx, FedOPT, FedKAD) on public datasets, and the reported accuracy and robustness gains stand independently of Theorem 1. The only circularity found is in the theoretical narrative around LPA: the claim that LPA bounds the deviation δ² is exactly the content of Assumption 5, and the proof does not derive this bound from the pruning mechanism. Additionally, there is a separate derivation gap unrelated to circularity: Eq. (27) bounds a weighted sum of per-client local Plug-in losses, while Eq. (28) claims a bound on the global gradient ∥∇L(θ^g_t)∥², with no comparison argument and no explicit definition of the global Plug-in objective L(θ^g). Self-citations in the paper are background references and are not load-bearing for the proposed framework or its evaluation. Overall, the circularity burden is mild and confined to the theoretical robustness guarantee, not the experimental contributions.

Assumptions & free parameters 2 free parameters · 6 assumptions · 1 invented entities

The central claim rests on a handful of free hyperparameters that are not reported, and on Assumption 5 which postulates the core benefit of the proposed LPA mechanism. These are the main entries the reader 'pays for upstream' beyond standard FL assumptions.

free parameters (2)

alpha = not reported
Balances task CE loss and KL distillation loss in Eq. 8; no value given in the experiments, so the reported results depend on an undisclosed choice.
v_h and v_l = not reported
Pruning fractions in Eq. 12 controlling how many extreme layer deviations are discarded each round; values are never given, yet they directly control the LPA aggregation.

assumptions (6)

domain assumption Assumption 1: local and plug-in losses are L-smooth
Standard in FL convergence proofs; assumed for all client models in Section 4.1.
domain assumption Assumption 2: gradients are bounded by G²
Standard bounded-gradient condition used in the convergence analysis.
domain assumption Assumption 3: bounded data heterogeneity (σ²)
Assumes local gradients differ from the global gradient by at most σ², a common non-IID bound.
domain assumption Assumption 4: unbiased stochastic gradients with bounded variance (σ_g²)
Standard mini-batch SGD assumption for both client and plug-in updates.
ad hoc to paper Assumption 5: LPA Effectiveness (bounded deviation δ² after pruning)
Assumes the pruning rule keeps deviations bounded, which is exactly the robustness property whose derivation is being claimed.
domain assumption Assumption 6: global loss bounded below
Standard lower-bound assumption for loss functions in optimization proofs.

invented entities (1)

Globally shared lightweight Plug-in model
purpose: Serves as a global knowledge carrier that each client distills from and into, enabling model heterogeneity without requiring identical local architectures.
This is an algorithmic construct proposed by the paper, not a physical entity. Its effectiveness is only demonstrated in the paper's own experiments and depends on the choice of architecture (CRNN-Lite).

how reviews work

0 comments

Cite this review

Pith. "Pith review of FedMLAC: Mutual Learning Driven Heterogeneous Federated Audio Classification." pith.science (2026). https://pith.science/paper/64KLIKCW

@misc{pith2026250610207,
  author       = {Pith},
  title        = {Pith review of: FedMLAC: Mutual Learning Driven Heterogeneous Federated Audio Classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/64KLIKCW}},
  note         = {Machine review of arXiv:2506.10207}
}

read the original abstract

Federated Learning (FL) offers a privacy-preserving framework for training audio classification (AC) models across decentralized clients without sharing raw data. However, Federated Audio Classification (FedAC) faces three major challenges: data heterogeneity, model heterogeneity, and data poisoning, which degrade performance in real-world settings. While existing methods often address these issues separately, a unified and robust solution remains underexplored. We propose FedMLAC, a mutual learning-based FL framework that tackles all three challenges simultaneously. Each client maintains a personalized local AC model and a lightweight, globally shared Plug-in model. These models interact via bidirectional knowledge distillation, enabling global knowledge sharing while adapting to local data distributions, thus addressing both data and model heterogeneity. To counter data poisoning, we introduce a Layer-wise Pruning Aggregation (LPA) strategy that filters anomalous Plug-in updates based on parameter deviations during aggregation. Extensive experiments on four diverse audio classification benchmarks, including both speech and non-speech tasks, show that FedMLAC consistently outperforms state-of-the-art baselines in classification accuracy and robustness to noisy data.

Figures

Figures reproduced from arXiv: 2506.10207 by the authors.

**Figure 2.** Performance comparison among FedKAD, FedMLAC without LPA, and FedM [PITH_FULL_IMAGE:figures/full_fig_p021_2.png] view at source ↗

**Figure 3.** Ablation comparison between FedMLAC and its variant without LPA and ML [PITH_FULL_IMAGE:figures/full_fig_p026_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 39 canonical work pages

[1]

X. Wang, Z. Su, J. Rekimoto, Y. Zhang, Watch your mouth: Silent speech recognition with depth sensing, in: Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems, 2024, pp. 1–15

work page 2024
[2]

G. Tu, T. Xie, B. Liang, H. Wang, R. Xu, Adaptive graph learning for multimodal conversational emotion detection, in: Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 38, 2024, pp. 19089– 19097

work page 2024
[3]

H. Xu, Y. Tian, H. Ren, X. Liu, A lightweight channel and time attention enhanced 1d cnn model for environmental sound classification, Expert Systems with Applications 249 (2024) 123768

work page 2024
[4]

J. Bai, A. Sajjanhar, Y. Xiang, X. Tong, S. Zeng, Fedewa: Federated learning with elastic weighted averaging, in: 2022 International Joint Conference on Neural Networks (IJCNN), IEEE, 2022, pp. 1–8

work page 2022
[5]

J. Bai, D. Wu, S. Zeng, Y. Zhao, Y. Qu, S. Yu, Non-iid free federated learning with fuzzy optimization for consumer electronics systems, IEEE Transactions on Consumer Electronics (2025)

work page 2025
[6]

H. Tong, Z. Yang, S. Wang, Y. Hu, O. Semiari, W. Saad, C. Yin, Fed- erated learning for audio semantic communication, Frontiers in commu- nications and networks 2 (2021) 734402

work page 2021
[7]

Ahmad, W

A. Ahmad, W. Luo, A. Robles-Kelly, Robust federated learning un- der statistical heterogeneity via hessian spectral decomposition, Pattern Recognition 141 (2023) 109635. 27

work page 2023
[8]

J. Bai, Y. Song, D. Wu, A. Sajjanhar, Y. Xiang, W. Zhou, X. Tao, Y. Li, Y. Li, A unified solution to diverse heterogeneities in one-shot federated learning, arXiv preprint arXiv:2410.21119 (2024)

work page Pith review arXiv 2024

Show all 44 references

[9]

D. Wu, J. Bai, Y. Song, J. Chen, W. Zhou, Y. Xiang, A. Sajjanhar, Fedinverse: Evaluating privacy leakage in federated learning, in: The twelfth international conference on learning representations, 2024

2024
[10]

Grollmisch, T

S. Grollmisch, T. K¨ ollmer, A. Yaroshchuk, H. Lukashevich, Federated semi-supervised learning for industrial sound analysis and keyword spot- ting, in: 2025 IEEE International Conference on Acoustics, Speech, and Signal Processing Workshops (ICASSPW), IEEE, 2025, pp. 1–5

2025
[11]

C. Xu, Z. Wu, F. Ge, Y. Zhi, Dynamic client selection and group- balanced personalization for data-imbalanced federated speech recog- nition, Electronics 14 (7) (2025) 1485

2025
[12]

Z. Lin, B. Zeng, H. Hu, Y. Huang, L. Xu, Z. Yao, Sase: Self-adaptive noise distribution network for speech enhancement with federated learning using heterogeneous data, Knowledge-Based Systems 266 (2023) 110396. doi:https://doi.org/10.1016/j.knosys.2023.110396. URL https://www...

2023
[13]

X. Chen, S. Yu, Y. Liu, et al., Federated learning for speech recognition: A review, IEEE Transactions on Neural Networks and Learning Systems 32 (8) (2021) 3510–3524

2021
[14]

Zhang, J

Q. Zhang, J. Cheng, J. Yang, et al., Distributed speech recognition with heterogeneous data, IEEE Transactions on Audio, Speech, and Language Processing 29 (2021) 270–284

2021
[15]

H. Li, Z. Cai, J. Wang, J. Tang, W. Ding, C.-T. Lin, Y. Shi, Fedtp: Federated learning by transformer personalization, IEEE transactions on neural networks and learning systems (2023)

2023
[16]

X. Kan, Y. Xiao, T.-J. Yang, N. Chen, R. Mathews, Parameter-efficient transfer learning under federated learning for automatic speech recogni- tion, arXiv preprint arXiv:2408.11873 (2024). 28

2024 arXiv
[17]

H. Xu, Z. Fan, X. Liu, Application of personalized federated learning methods to environmental sound classification: A comparative study, Engineering Applications of Artificial Intelligence 135 (2024) 108760

2024
[18]

Zhang, Y

D. Zhang, Y. Yu, C. Li, J. Dong, D. Yu, Federated incremental named entity recognition, IEEE Transactions on Audio, Speech and Language Processing (2025)

2025
[19]

M. N. Ali, D. Falavigna, A. Brutti, Efl-peft: A communication effi- cient federated learning framework using peft sparsification for asr, in: ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), IEEE, 2025, pp. 1–5

2025
[20]

N. S. Senol, M. Baza, A. Rasheed, M. Alsabaan, Privacy-preserving detection of tampered radio-frequency transmissions utilizing federated learning in lora networks, Sensors 24 (22) (2024) 7336

2024
[21]

M. A. ALMAIAH, A. ALI, R. SHISHAKLY, T. ALKHDOUR, A. LUTFI, M. ALRA W AD, A novel federated-learning based adversarial framework for audio-visual speech enhancement, Journal Of Theoretical And Applied Information Technology 102 (4) (2024)

2024
[22]

A. K. Bhuyan, H. Dutta, S. Biswas, Unsupervised speaker diarization in distributed iot networks using federated learning, IEEE Transactions on Emerging Topics in Computational Intelligence (2024)

2024
[23]

Benazir, F

A. Benazir, F. X. Lin, Privacy-preserving edge speech understanding with tiny foundation models, arXiv preprint arXiv:2502.01649 (2025)

2025
[24]

Hsu, C.-P

W.-T. Hsu, C.-P. Chen, Y.-S. Lin, C.-C. Lee, A cluster-based personal- ized federated learning strategy for end-to-end asr of dementia patients, in: Proceedings of Interspeech 2024, 2024, pp. 2450–2454

2024
[25]

Lee, J.-H

C.-W. Lee, J.-H. Lee, J.-H. Chang, Language model personalization for speech recognition: A clustered federated learning approach with adaptive weight average, IEEE Signal Processing Letters (2024)

2024
[26]

Leroy, A

D. Leroy, A. Coucke, T. Lavril, T. Gisselbrecht, J. Dureau, Federated learning for keyword spotting, in: ICASSP 2019-2019 IEEE interna- tional conference on acoustics, speech and signal processing (ICASSP), IEEE, 2019, pp. 6341–6345. 29

2019
[27]

Y. Du, Z. Zhang, L. Yue, X. Huang, Y. Zhang, T. Xu, L. Xu, E. Chen, Communication-efficient personalized federated learning for speech-to- text tasks, in: ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), IEEE, 2024, pp. 10001–10005

2024
[28]

Amiri, H

A. Amiri, H. Khosravi, et al., Federated learning with heterogeneous clients: A study on speech recognition, in: 2021 IEEE International Con- ference on Acoustics, Speech and Signal Processing (ICASSP), IEEE, 2021, pp. 3015–3019

2021
[29]

H. Zhu, J. Wang, G. Cheng, P. Zhang, Y. Yan, Decoupled federated learning for asr with non-iid data, in: Proceedings of Interspeech 2022, 2022, pp. 2628–2632

2022
[30]

Feng, C.-C

M. Feng, C.-C. Kao, Q. Tang, M. Sun, V. Rozgic, S. Matsoukas, C. Wang, Federated self-supervised learning for acoustic event classifica- tion, in: ICASSP 2022-2022 IEEE International Conference on Acous- tics, Speech and Signal Processing (ICASSP), IEEE, 2022, pp. 481–485

2022
[31]

Tsouvalas, A

V. Tsouvalas, A. Saeed, T. Ozcelebi, N. Meratnia, Labeling chaos to learning harmony: Federated learning with noisy labels, ACM Transac- tions on Intelligent Systems and Technology 15 (2) (2024) 1–26

2024
[32]

J. Jia, J. Mahadeokar, W. Zheng, Y. Shangguan, O. Kalinli, F. Seide, Federated domain adaptation for asr with full self-supervision, arXiv preprint arXiv:2203.15966 (2022)

2022 arXiv
[33]

Tsouvalas, A

V. Tsouvalas, A. Saeed, T. Ozcelebi, Federated self-training for data- efficient audio recognition, in: ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), IEEE, 2022, pp. 476–480

2022
[34]

H. Shi, V. Radu, P. Yang, Distributed training for speech recogni- tion using local knowledge aggregation and knowledge distillation in heterogeneous systems, in: Proceedings of the 3rd Workshop on Ma- chine Learning and Systems (EuroMLSys ’23), ACM, 2023, pp. 64–70. doi:10.11...

2023
[35]

McMahan, E

B. McMahan, E. Moore, D. Ramage, B. A. y Arcas, Communication- efficient learning of deep networks from decentralized data, in: Artificial Intelligence and Statistics, PMLR, 2017, pp. 1273–1282

2017
[36]

Zhang, T

Y. Zhang, T. Xiang, T. M. Hospedales, H. Lu, Deep mutual learning, in: Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 4320–4328

2018
[37]

Warden, Speech commands: A dataset for limited-vocabulary speech recognition, arXiv preprint arXiv:1804.03209 (2018)

P. Warden, Speech commands: A dataset for limited-vocabulary speech recognition, arXiv preprint arXiv:1804.03209 (2018)

2018 arXiv
[38]

Salamon, C

J. Salamon, C. Jacoby, J. P. Bello, A dataset and taxonomy for urban sound research, in: Proceedings of the 22nd ACM international confer- ence on Multimedia, 2014, pp. 1041–1044

2014
[39]

Busso, M

C. Busso, M. Bulut, C.-C. Lee, A. Kazemzadeh, E. Mower, S. Kim, J. N. Chang, S. Lee, S. S. Narayanan, Iemocap: Interactive emotional dyadic motion capture database, Language resources and evaluation 42 (2008) 335–359

2008
[40]

H. Cao, D. G. Cooper, M. K. Keutmann, R. C. Gur, A. Nenkova, R. Verma, Crema-d: Crowd-sourced emotional multimodal actors dataset, IEEE transactions on affective computing 5 (4) (2014) 377– 390

2014
[41]

Zhang, T

T. Zhang, T. Feng, S. Alam, S. Lee, M. Zhang, S. S. Narayanan, S. Aves- timehr, Fedaudio: A federated learning benchmark for audio tasks, in: ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), IEEE, 2023, pp. 1–5

2023
[42]

Chen, W.-L

H.-Y. Chen, W.-L. Chao, Fedbe: Making bayesian model ensemble ap- plicable to federated learning, arXiv preprint arXiv:2009.01974 (2020)

2020 arXiv
[43]

T. Li, A. Sahu, A. Talwalkar, V. Smith, Federated optimization in het- erogeneous networks, IEEE Transactions on Signal Processing (2020)

2020
[44]

Reddi, Z

S. Reddi, Z. Charles, M. Zaheer, Z. Garrett, K. Rush, J. Koneˇ cn` y, S. Kumar, H. B. McMahan, Adaptive federated optimization, arXiv preprint arXiv:2003.00295 (2020). 31

2020 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

X. Wang, Z. Su, J. Rekimoto, Y. Zhang, Watch your mouth: Silent speech recognition with depth sensing, in: Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems, 2024, pp. 1–15

work page 2024

[2] [2]

G. Tu, T. Xie, B. Liang, H. Wang, R. Xu, Adaptive graph learning for multimodal conversational emotion detection, in: Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 38, 2024, pp. 19089– 19097

work page 2024

[3] [3]

H. Xu, Y. Tian, H. Ren, X. Liu, A lightweight channel and time attention enhanced 1d cnn model for environmental sound classification, Expert Systems with Applications 249 (2024) 123768

work page 2024

[4] [4]

J. Bai, A. Sajjanhar, Y. Xiang, X. Tong, S. Zeng, Fedewa: Federated learning with elastic weighted averaging, in: 2022 International Joint Conference on Neural Networks (IJCNN), IEEE, 2022, pp. 1–8

work page 2022

[5] [5]

J. Bai, D. Wu, S. Zeng, Y. Zhao, Y. Qu, S. Yu, Non-iid free federated learning with fuzzy optimization for consumer electronics systems, IEEE Transactions on Consumer Electronics (2025)

work page 2025

[6] [6]

H. Tong, Z. Yang, S. Wang, Y. Hu, O. Semiari, W. Saad, C. Yin, Fed- erated learning for audio semantic communication, Frontiers in commu- nications and networks 2 (2021) 734402

work page 2021

[7] [7]

Ahmad, W

A. Ahmad, W. Luo, A. Robles-Kelly, Robust federated learning un- der statistical heterogeneity via hessian spectral decomposition, Pattern Recognition 141 (2023) 109635. 27

work page 2023

[8] [8]

J. Bai, Y. Song, D. Wu, A. Sajjanhar, Y. Xiang, W. Zhou, X. Tao, Y. Li, Y. Li, A unified solution to diverse heterogeneities in one-shot federated learning, arXiv preprint arXiv:2410.21119 (2024)

work page Pith review arXiv 2024