Multi-Phonation Graph Learning with Self-Supervised Speech Embeddings for ALS Detection and Progression Prediction

Behrad TaghiBeyglou; Ervin Sejdic; Fatemeh Bagheri

REVIEW 3 major objections 6 minor 29 references

Multi-Phonation Graph Learning with Self-Supervised Speech Embeddings for ALS Detection and Progression Prediction

T0 review · 3 major / 6 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read Fusing a patient's voice clips into one graph improves ALS severity and progression prediction.

desk verdict Solid, clearly written benchmark with a real new configuration, but the headline margin is built on best-of-many validation selection; treat the 0.73/0.69 as promising, not proven. read the letter →

arxiv 2607.25284 v1 pith:FADKXDQO submitted 2026-07-28 eess.AS cs.SD

Behrad TaghiBeyglou , Fatemeh Bagheri , Ervin Sejdic This is my paper · ORCID

classification eess.AScs.SD

keywords ALSdysarthriaseverityself-supervisedspeechembeddingsHuBERTgraphneuralnetworksGINprogressionpredictionmulti-phonationfusion

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 a patient's multiple short phonation recordings—five sustained vowels and three diadochokinetic syllable repetitions—can be treated as one connected structure rather than as independent recordings. It builds a single k-nearest-neighbor graph per patient in the embedding space of frozen self-supervised speech models, then classifies the whole graph with a graph neural network. On the official validation split, the best configuration (HuBERT embeddings with a GIN aggregator) reaches macro-F1 of 0.73 for five-class dysarthria severity and 0.69 for four-class ALSFRS-R progression prediction, above challenge baselines of 0.61 and 0.58. A sympathetic reader cares because this is a step toward scalable, low-resource voice biomarkers: if the gain is real, combining multiple brief voice tasks through graph-based message passing captures complementary cues that per-recording pipelines miss.

What carries the argument

The central mechanism is a per-subject k-nearest-neighbor graph built in the embedding space of frozen self-supervised speech encoders. Nodes are mean-pooled 768-dimensional embeddings of 2-second chunks; edges are symmetrized kNN connections weighted by cosine similarity. A GIN (graph isomorphism network with sum aggregation and MLP updates) performs message passing over this graph, and global mean pooling produces the graph-level representation fed to a classifier—the sum aggregation is credited with preserving sparse informative cues.

What would settle it

Take the best HuBERT+GIN pipeline and destroy the graph topology—randomly rewire the edges or replace the kNN graph with a graph over shuffled chunk order while keeping the same node features. If validation macro-F1 does not materially drop, the claimed benefit of multi-phonation fusion through message passing is not real. A second check: a per-subject classifier that simply averages the same HuBERT chunk embeddings with no edges at all; if it matches 0.73, the GNN is superfluous.

Watch

Extended reading notes

Core claim

The authors claim that a subject-level kNN graph over SSL chunk embeddings, classified by a graph neural network, outperforms per-recording baselines on both SAND tasks. Each 2-second chunk of a patient's recordings becomes a node; edges connect chunks with the highest cosine similarity, and a GIN with sum aggregation pools all nodes into a subject-level prediction. The paper reports HuBERT+GIN achieving macro-F1 of 0.73 for dysarthria severity and 0.69 for ALSFRS-R progression on the official validation split, versus 0.61 and 0.58 for the baselines. The authors argue that this demonstrates the value of fusing evidence across phonation types and temporal chunks.

Load-bearing premise

The load-bearing premise is that the kNN graph's edges, computed by cosine similarity over frozen speech embeddings, connect chunks that are similar because of clinically relevant dysarthria cues rather than because of recording-channel artifacts or the artificial repetition introduced by tiling short recordings; the paper itself acknowledges this limitation.

Editorial extensions

If this is right

If the validation results hold on the hidden test set, per-subject graph fusion becomes a strong default design for speech biomarker tasks with multiple recordings per person.
The HuBERT+GIN pairing working on both severity and progression suggests the benefit is not specific to one task and may transfer to other neurodegenerative speech assessments.
Graph aggregation naturally handles missing or noisy recordings because the graph is built over whatever chunks exist per subject.
Frozen SSL front-ends require no task-specific labels, fitting low-resource clinical settings where labeled ALS speech is scarce.
The consistent advantage of GIN over mean-based models points to sparse, distributed acoustic cues being the information that multi-phonation fusion exploits.

Reading between the lines

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

The reported margin is likely to shrink on the held-out test set: the paper's own evidence is limited to the validation split, and test-set generalization is explicitly unverified.
The paper's Discussion concedes that the unsupervised kNN edges may connect chunks by nuisance similarity such as channel characteristics rather than clinically meaningful acoustic proximity, which is the main threat to the interpretation of the gains.
Because short recordings are tiled to reach 20 seconds, many of the 80 chunks per subject are near-duplicates; running the same pipeline with deduplicated chunks or edges restricted to cross-recording neighbors would clarify whether the gains come from genuine cross-phonation fusion or from amplifying artificial self-similarity.
A stronger non-graph baseline—for example, a transformer or set-pooling network over the same chunk embeddings—would isolate the value of message passing from the value of simply having more acoustic context per subject.

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

One thing to know: this is a clean, competent benchmark of SSL embeddings + kNN graph + GNN for ALS detection and progression on the SAND challenge. The new piece is applying the already-established graph-based pathological speech recipe to ALS multi-phonation, and the specific HuBERT+GIN configuration gets validation mF1 0.73 (Task 1) and 0.69 (Task 2), well above the fixed SAND baselines (0.61/0.58). The catch: those numbers are selected as the best among a grid of 4 SSL encoders x 5 GNNs x k/d/L/dropout/LR on the same validation split used for the comparison, so the margin is not as trustworthy as the abstract implies.

What the paper does well: the methods are described clearly, the experiments are systematic, and the authors explicitly list the main limitations—tiling/truncation artifacts, unsupervised kNN possibly capturing nuisance similarity, and no test-set confirmation. That candor is real.

The soft spots are proportional. The selection protocol is the biggest: Section 2.6 says hyperparameters are chosen by highest validation mF1, and Table 1 confirms each cell is 'selected by highest validation mF1.' So the headline is best-of-many versus a single baseline. That inflates the apparent advantage. Also, Section 3.3 says 'consistent margin across all our configurations,' but Table 1 shows several cells below the baselines (e.g., UniSpeech-SAT+GAT 0.52 Task 1; Data2Vec+GCN 0.49 Task 2). So the robust conclusion is not that graph aggregation broadly helps; it's that one configuration, chosen with validation labels, did. The paper's own Discussion concedes the kNN 'may connect segments by nuisance similarity,' and the tiling creates near-duplicate chunks that could dominate the graph, so the mechanism behind the gains is under-supported.

Still, the idea is plausible and the claim is testable. With a pre-registered or fixed protocol and a test-server submission, this could be a solid contribution. I would not cite the headline numbers as established, but I would engage with it as a useful data point on validation-selection pitfalls in clinical speech ML.

Recommendation: send to peer review, but demand a corrected comparison (single model or proper nested CV) and ideally a test-set result before acceptance.

Referee Report

3 major / 6 minor

Summary. This manuscript proposes a subject-level graph learning framework for ALS assessment on the SAND challenge. Each subject's multiple phonation/DDK recordings are resampled, tiled/truncated to a fixed 20 s duration, segmented into 2 s chunks, embedded with frozen self-supervised speech models, and represented as a k-nearest-neighbor graph over chunk embeddings. A graph neural network with global mean pooling performs 5-class dysarthria severity classification (Task 1) and 4-class ALSFRS-R progression prediction (Task 2). The authors benchmark four SSL front-ends (wav2vec 2.0, HuBERT, data2vec-audio, UniSpeech-SAT) across five GNN architectures (GCN, ResGCN, GAT, GraphSAGE, GIN) and report best validation macro-F1 values of 0.73 (Task 1) and 0.69 (Task 2) for HuBERT+GIN, compared with SAND validation baselines of 0.61 and 0.58.

Significance. If the headline results hold under a properly controlled evaluation protocol, the idea of constructing subject-level graphs from frozen SSL embeddings is a plausible and potentially useful direction for low-resource clinical speech analysis. The paper covers a broad configuration sweep and is transparent about several limitations (cross-lingual mismatch, tiling artifacts, unsupervised graph topology, and lack of test-set confirmation). However, the central empirical superiority claim is not currently established because of the way the validation split is used for model selection, the lack of an identical statistical footing for the baseline comparison, and the absence of uncertainty quantification. The significance is therefore conditional; the contribution is more methodological than a demonstrated performance breakthrough until these evaluation issues are addressed.

major comments (3)

[2.6 / Table 1 / 3.3] The reported superiority of HuBERT+GIN is not established by the current evaluation protocol. Section 2.6 states that hyperparameters are selected by 10-fold CV over all subjects, but Table 1's caption says each configuration is "selected by highest validation mF1." If the latter is the case, the headline 0.73/0.69 are maxima over a grid of 4 SSL encoders × 5 GNNs × k ∈ {1,3,5,10} × d ∈ {128,256} × L ∈ {2,3} × dropout ∈ {0.3,0.5} × learning rate ∈ {1e-3,3e-4}, all evaluated on the same 53-subject official validation split used for the baseline comparison. Best-of-many selection on a small validation set can substantially inflate apparent margins. The claim in Section 3.3 of a "consistent margin across all our configurations" is also contradicted by Table 1, where e.g. UniSpeech-SAT+GAT gives 0.52 (Task 1) and Data2Vec+GCN gives 0.49 (Task 2), below the 0.61/0.58 baselines. Please clarify
[3.3 / Table 2] The comparison to SAND baselines is statistically under-specified. The baselines are given as point values (0.61/0.58) without confidence intervals or standard deviations, and Table 2 labels them "Val" while the text in Section 3.3 refers to "SAND leaderboard's top scores" on the held-out test set. For the conclusion that the proposed method outperforms the baselines, the comparison must be on identical splits, identical metric definitions, and include a measure of uncertainty (e.g., bootstrap CIs, multi-seed variance, or a significance test). Without this, a 0.12 mF1 gap on 53 subjects is not sufficient evidence, especially when the comparison is between a best-of-many selected configuration and a fixed baseline.
[2.2 / 2.4 / 4] The graph construction may be dominated by tiling artifacts. Eq. (1) builds kNN edges by cosine similarity among chunk embeddings; because Section 2.2 creates exactly 10 segments per 20 s clip by repeating shorter recordings, many nodes are near-duplicates of the same audio. Such duplicated nodes will be mutually nearest neighbors and can distort both the graph topology and the message-passing readout. The manuscript acknowledges this possibility in the Limitations but does not test its impact. Please quantify the fraction of edges connecting duplicated chunks and provide an ablation that removes duplicated chunks or uses variable-length segment counts; this is necessary to support the interpretation that the gains come from multi-phonation fusion rather than self-similarity.

minor comments (6)

[Fig. 1] The Task 2 output classes are labeled with the same dysarthria severity classes as Task 1; this appears to be a copy-paste error and should be corrected to the ALSFRS-R progression classes.
[2.6] The two-stage protocol (10-fold CV for hyperparameter selection, then retrain on official training split) should specify whether the official validation subjects were included in the 10-fold CV; if so, this must be disclosed because it means validation labels were used for model selection.
[2.4] The symbol E is used both for the node feature matrix and for the edge set in Eq. (1); use different symbols to avoid ambiguity.
[2.1] The class definitions for Task 2 (progression) are not given; specify the ALSFRS-R thresholds or grouping used by SAND. Also state the original recording durations before tiling.
[References] Reference [15] is a website; include the official challenge description or technical report with details on the dataset and baselines, and cite the baseline methods (ViT/PART) properly.
[Reproducibility] No link to code or trained models is provided; to support reproducibility, please include a public repository with the graph construction and training scripts.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the pipeline is self-contained, with validation-set selection as a soundness caveat, not a circular step.

full rationale

After walking the claimed derivation chain, I find no circular step of the kind defined here. Task 1 uses contemporaneous speech labels and Task 2 predicts future ALSFRS-R from early recordings; the target labels are never used to construct the kNN graph, to set the frozen SSL embeddings, or to define edge weights (Eq. 1 is purely a cosine-similarity function of the input embeddings). The GNN classifiers are trained with standard losses, and the reported validation scores are empirical measurements, not quantities forced to equal the labels by construction. References such as [15] are external challenge baselines; there is no load-bearing self-citation, no imported uniqueness theorem, and no ansatz smuggled in via citation. The manuscript's own limitations — tiling artifacts, nuisance similarity in the unsupervised kNN topology, and the need to confirm test-set generalization — are candid soundness caveats, not admissions of circularity. The main methodological concern is that Table 1 states 'Each configuration per SSL × GNN for Task 1 and Task 2 is selected by highest validation mF1' and Section 2.6 runs early stopping on the validation split, so the headline numbers are selected on the same split used for the baseline comparison. That is a selection-bias/statistical-inference weakness, not a reduction of the result to its inputs; the model is still trained on the training split and the labels do not enter the graph construction. Therefore the honest finding is no significant circularity (score 0).

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

The central claim rests on model-selection choices (k, d, L, dropout, learning rate, and ultimately the SSL×GNN pairing) tuned on the official validation split; these are free parameters. It also assumes that frozen English SSL embeddings transfer to Italian pathological speech, that the kNN graph captures clinically meaningful structure, and that tiling/truncation does not distort the signal. No new physical or conceptual entities are introduced.

free parameters (5)

kNN neighborhood size k = 1 (Task 1 HuBERT+GIN best), 10 (Task 2 HuBERT+GIN best)
Chosen by grid search over {1,3,5,10} on the official validation split (Section 2.4, Table 1); headline scores are conditional on this choice.
hidden dimension d = 128 (Task 1 best), 256 (Task 2 best)
Grid-searched over {128,256} per GNN/SSL combination and selected by validation mF1 (Section 2.5, Table 1).
number of message-passing layers L = not reported per best config; searched over {2,3}
Grid-searched but the final L for the headline HuBERT+GIN configurations is not stated in Table 1.
dropout probability = not reported per best config; searched over {0.3,0.5}
Grid-searched (Section 2.5) and selected on validation, but not reported per configuration.
learning rate = not reported per best config; searched over {1e-3,3e-4}
Grid-searched (Section 2.6) and selected on validation, but not reported per configuration.

assumptions (5)

domain assumption Frozen SSL embeddings pretrained on English speech transfer to Italian pathological speech.
Used in Section 2.3 to featurize Italian SAND audio; the authors acknowledge cross-lingual mismatch as a limitation in the Discussion.
domain assumption Cosine-similarity kNN over embeddings captures clinically meaningful acoustic proximity.
Core graph-construction premise, Eq. (1) in Section 2.4; authors concede the kNN may connect segments by nuisance similarity.
ad hoc to paper Tiling shorter recordings to a fixed 20 s duration does not distort the acoustic signal.
Section 2.2 preprocessing repeats/truncates recordings; authors acknowledge artificial repetition may bias neighborhood construction.
domain assumption Message passing over the subject graph can propagate dysarthria cues across heterogeneous phonation types.
The proposed mechanism in Sections 2.5 and 4; not separately validated, e.g., via an ablation on edge structure.
domain assumption The official SAND validation split is an appropriate benchmark and the reported SAND baseline scores are protocol-comparable.
Section 3.3 compares against baseline numbers from [15] on the same validation split, but the baselines were not tuned on that split as the present models were.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Multi-Phonation Graph Learning with Self-Supervised Speech Embeddings for ALS Detection and Progression Prediction." pith.science (2026). https://pith.science/paper/FADKXDQO

@misc{pith2026260725284,
  author       = {Pith},
  title        = {Pith review of: Multi-Phonation Graph Learning with Self-Supervised Speech Embeddings for ALS Detection and Progression Prediction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FADKXDQO}},
  note         = {Machine review of arXiv:2607.25284}
}

abstract

Amyotrophic lateral sclerosis (ALS) progressively impairs speech motor control, making acoustic analysis a promising biomarker for severity and progression estimation. We propose a subject-level graph framework that aggregates multiple phonation recordings into a unique k-nearest-neighbor graph built from pretrained SSL embeddings of 2s segments. We compare four SSL front-ends (wav2vec 2.0, HuBERT, data2vec-audio, and UniSpeech-SAT) and five graph neural networks (GCN, residual GCN, GAT, GraphSAGE, and GIN) on the SAND dataset tasks (339 participants: 205 ALS, 134 control): 5-class dysarthria severity and 4-class ALSFRS-R progression prediction. On the official validation set, the best configuration (HuBERT+GIN) achieves macro-F$_1$ of 0.73 for Task 1 and 0.69 for Task 2, outperforming SAND validation baselines (0.61 and 0.58). These results highlight the potential of combining GNNs with pretrained cross-lingual speech representations for low-resource ALS detection and progression monitoring.

Figures

Figures reproduced from arXiv: 2607.25284 by the authors.

**Figure 1.** Proposed pipeline. For each patient, multiple recording segments are independently encoded by a frozen SSL model, assembled into a kNN graph, and classified via a GNN with global pooling. 3. Results 3.1. Task 1: Dysarthria Severity Classification Although Task 1 involves a larger label space (5 classes), it is generally easier than Task 2 because it relies on contemporaneous acoustic cues of dysarthria rather than … view at source ↗

**Figure 2.** Best 10-fold cross-validation mF1 for both tasks per SSL × GNN combination (error bars denote CV σ). compared to the SAND leaderboard’s top scores of 0.61 and 0.58 on the held-out test set. We caution that a direct comparison is confounded by the partition difference, our numbers are computed on the internal validation split rather than the final test set. Nonetheless, the comparison with the baseline of mF1 of 0.6… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 4 linked inside Pith

[1]

Introduction Amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disease that progressively impairs motor function, frequently affecting speech early in the disease course [ 1]. Dysarthria- related changes in sustained vowel phonation and diadochokine- sis (DDK) are clinically informative and can be captured non- invasively with brief voice t...
[2]

Dataset The SAND Challenge dataset [15, 2, 3] comprises 2,712 voice recordings from 339 Italian speakers, including 205 ALS pa- tients and 134 healthy controls

Methods 2.1. Dataset The SAND Challenge dataset [15, 2, 3] comprises 2,712 voice recordings from 339 Italian speakers, including 205 ALS pa- tients and 134 healthy controls. Each subject provides five sustained vowel phonations (/a/, /e/, /i/, /o/, /u/ ) and three DDK syllables with frequent repetitions ( /pa/, /ta/, /ka/ ). The official split includes 21...

arXiv 2026
[3]

Results 3.1. Task 1: Dysarthria Severity Classification Although Task 1 involves a larger label space (5 classes), it is generally easier than Task 2 because it relies on contempora- neous acoustic cues of dysarthria rather than forecasting future functional decline, acoustic markers of which can be weaker and confounded by inter-subject variability and h...
[4]

Discussion We proposed a subject-level graph learning pipeline that converts multiple short speech segments per speaker into a single kNN graph in SSL-embedding space and performs graph classification using standard GNN backbones. Across both SAND tasks, the same pairing, HuBERT embeddings with a GIN classifier, con- sistently delivered the strongest vali...
[5]

Conclusion We introduced a simple but effective way to turn multiple short phonation recordings from a single speaker into a uni- fied, subject-level representation; a kNN graph built in the space of frozen SSL embeddings and classified with a graph neural network. Across an extensive sweep of different configurations per task on the SAND dataset, the sam...
[6]

Generative AI Use Disclosure The authors take full responsibility for the accuracy, originality, and integrity of the final work and affirm that generative AI was used solely for language refinement and drafting assistance
[7]

Self-supervised speech representation learning: A review,

A. Mohamed, H.-y. Lee, L. Borgholt, J. D. Havtorn, J. Edin, C. Igel, K. Kirchhoff, S.-W. Li, K. Livescu, L. Maaløe, T. N. Sainath, and S. Watanabe, “Self-supervised speech representation learning: A review,”arXiv preprint arXiv:2205.10643, 2022

arXiv 2022
[8]

Amyotrophic lateral sclerosis,

M. C. Kiernan, S. Vucic, B. C. Cheah, M. R. Turner, A. Eisen, O. Hardiman, J. R. Burrell, and M. C. Zoing, “Amyotrophic lateral sclerosis,”The Lancet, vol. 377, no. 9769, pp. 942–955, 2011

2011

Show all 29 references

[9]

V ox4health: Prelimi- nary results of a pilot study for the evaluation of a mobile voice screening application,

L. Verde, G. De Pietro, and G. Sannino, “V ox4health: Prelimi- nary results of a pilot study for the evaluation of a mobile voice screening application,” inInternational Symposium on Ambient Intelligence. Springer, 2016, pp. 131–140

2016
[10]

V oice signals database of als patients with different dysarthria severity and healthy controls,

R. Dubbioso, M. Spisto, L. Verde, V . V . Iuzzolino, G. Senerchia, E. Salvatore, G. De Pietro, I. De Falco, and G. Sannino, “V oice signals database of als patients with different dysarthria severity and healthy controls,”Scientific Data, vol. 11, no. 1, p. 800, 2024

2024
[11]

Detecting bulbar involvement in patients with amyotrophic lateral sclerosis based on phonatory and time-frequency features,

A. Tena, F. Clarià, F. Solsona, and M. Povedano, “Detecting bulbar involvement in patients with amyotrophic lateral sclerosis based on phonatory and time-frequency features,”Sensors, vol. 22, no. 3, p. 1137, 2022

2022
[12]

Clinical assessment and interpretation of dysarthria in ALS using attention based deep learning AI models,

M. Merler, C. Agurto, J. Peller, E. Roitberg, A. Taitz, M. A. Tre- visan, I. Navar, J. D. Berry, E. Fraenkel, L. W. Ostrow, G. A. Cecchi, and R. Norel, “Clinical assessment and interpretation of dysarthria in ALS using attention based deep learning AI models,” npj Digital Medi...

2025
[13]

Automated dysarthria severity classifi- cation: A study on acoustic features and deep learning techniques,

A. A. Joshy and R. Rajan, “Automated dysarthria severity classifi- cation: A study on acoustic features and deep learning techniques,” IEEE Transactions on Neural Systems and Rehabilitation Engi- neering, vol. 30, pp. 1147–1157, 2022

2022
[14]

UniSpeech-SAT: Universal speech repre- sentation learning with speaker aware pre-training,

S. Chen, Y . Wu, C. Wang, Z. Chen, Z. Chen, S. Liu, J. Wu, Y . Qian, F. Wei, J. Li, and X. Yu, “UniSpeech-SAT: Universal speech repre- sentation learning with speaker aware pre-training,” inProceedings of the IEEE International Conference on Acoustics, Speech and Signal Proces...

2022
[15]

and consistent margin across all our configurations suggests that the graph-based multi-segment modelling provides strong representation that conventional per-recording approaches do not exploit
[16]

SUPERB: Speech processing universal performance benchmark,

S.-w. Yang, P.-H. Chi, Y .-S. Chuang, C.-I. J. Lai, K. Lakhotia, Y . Y . Lin, A. T. Liu, J. Shi, X. Chang, G.-T. Lin, T.-H. Huang, W.-C. Tseng, T.-Q. Lin, K.-T. Lee, D.-R. Liu, Z. Huang, S. Dong, S.-W. Li, S. Watanabe, A. Mohamed, and H.-y. Lee, “SUPERB: Speech processing univ...

2021 arXiv
[17]

Graph-based semi-supervised learning for pathological speech detection,

W. Cai, M. Wu, L. Hou, and J. Tao, “Graph-based semi-supervised learning for pathological speech detection,” inIEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2022

2022
[18]

Graph neural networks for parkinson’s disease detection,

S. A. Sheikh, Y . Kaloga, M. Sahidullah, and I. Kodrasi, “Graph neural networks for parkinson’s disease detection,” inICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2025, pp. 1–5

2025
[19]

wav2vec 2.0: A framework for self-supervised learning of speech representations,

A. Baevski, H. Zhou, A. Mohamed, and M. Auli, “wav2vec 2.0: A framework for self-supervised learning of speech representations,” arXiv preprint arXiv:2006.11477, 2020

2006 arXiv
[20]

HuBERT: Self-supervised speech representa- tion learning by masked prediction of hidden units,

W.-N. Hsu, B. Bolte, Y .-H. H. Tsai, K. Lakhotia, R. Salakhutdinov, and A. Mohamed, “HuBERT: Self-supervised speech representa- tion learning by masked prediction of hidden units,”IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 29, pp. 3451–3460, 2021

2021
[21]

data2vec: A general framework for self-supervised learning in speech, vision and language,

A. Baevski, W.-N. Hsu, Q. Xu, A. Babu, J. Gu, and M. Auli, “data2vec: A general framework for self-supervised learning in speech, vision and language,” inInternational Conference on Ma- chine Learning (ICML), 2022, pp. 1298–1312

2022
[22]

Speech analysis for neurodegener- ative diseases (SAND) challenge,

SAND Organizing Committee, “Speech analysis for neurodegener- ative diseases (SAND) challenge,” 2025, grand Challenge at IEEE ICASSP 2026, Barcelona, Spain. https://www.sand.icar.cnr.it/

2025
[23]

The ALSFRS-R: A revised ALS func- tional rating scale that incorporates assessments of respiratory function,

J. M. Cedarbaum, N. Stambler, E. Malta, C. Fuller, D. Hilt, B. Thur- mond, and A. Nakanishi, “The ALSFRS-R: A revised ALS func- tional rating scale that incorporates assessments of respiratory function,”Journal of the Neurological Sciences, vol. 169, no. 1-2, pp. 13–21, 1999

1999
[24]

Semi-supervised classification with graph convolutional networks,

T. N. Kipf and M. Welling, “Semi-supervised classification with graph convolutional networks,” inInternational Conference on Learning Representations (ICLR), 2017. [Online]. Available: https://openreview.net/forum?id=SJU4ayYgl

2017
[25]

Graph attention networks,

P. Veliˇckovi´c, G. Cucurull, A. Casanova, A. Romero, P. Liò, and Y . Bengio, “Graph attention networks,” inInternational Conference on Learning Representations (ICLR), 2018

2018
[26]

Inductive representation learning on large graphs,

W. Hamilton, Z. Ying, and J. Leskovec, “Inductive representation learning on large graphs,” inAdvances in Neural Information Processing Systems (NeurIPS),
[28]

How powerful are graph neural networks?

K. Xu, W. Hu, J. Leskovec, and S. Jegelka, “How powerful are graph neural networks?” inInternational Conference on Learning Representations (ICLR), 2019. [Online]. Available: https://openreview.net/forum?id=ryGs6iA5Km

2019
[29]

Acoustic analysis of dysarthria profile in ALS patients,

B. Tomik, J. Krupinski, L. Glodzik-Sobanska, M. Bala-Slodowska, W. Wszolek, M. Kusiak, and A. Lechwacka, “Acoustic analysis of dysarthria profile in ALS patients,”Journal of the Neurological Sciences, vol. 169, no. 1–2, pp. 35–42, 1999

1999
[2017]

Available: https://papers.nips.cc/paper/ 6703-inductive-representation-learning-on-large-graphs

[Online]. Available: https://papers.nips.cc/paper/ 6703-inductive-representation-learning-on-large-graphs

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.

[1] [1]

Introduction Amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disease that progressively impairs motor function, frequently affecting speech early in the disease course [ 1]. Dysarthria- related changes in sustained vowel phonation and diadochokine- sis (DDK) are clinically informative and can be captured non- invasively with brief voice t...

[2] [2]

Dataset The SAND Challenge dataset [15, 2, 3] comprises 2,712 voice recordings from 339 Italian speakers, including 205 ALS pa- tients and 134 healthy controls

Methods 2.1. Dataset The SAND Challenge dataset [15, 2, 3] comprises 2,712 voice recordings from 339 Italian speakers, including 205 ALS pa- tients and 134 healthy controls. Each subject provides five sustained vowel phonations (/a/, /e/, /i/, /o/, /u/ ) and three DDK syllables with frequent repetitions ( /pa/, /ta/, /ka/ ). The official split includes 21...

arXiv 2026

[3] [3]

Results 3.1. Task 1: Dysarthria Severity Classification Although Task 1 involves a larger label space (5 classes), it is generally easier than Task 2 because it relies on contempora- neous acoustic cues of dysarthria rather than forecasting future functional decline, acoustic markers of which can be weaker and confounded by inter-subject variability and h...

[4] [4]

Discussion We proposed a subject-level graph learning pipeline that converts multiple short speech segments per speaker into a single kNN graph in SSL-embedding space and performs graph classification using standard GNN backbones. Across both SAND tasks, the same pairing, HuBERT embeddings with a GIN classifier, con- sistently delivered the strongest vali...

[5] [5]

Conclusion We introduced a simple but effective way to turn multiple short phonation recordings from a single speaker into a uni- fied, subject-level representation; a kNN graph built in the space of frozen SSL embeddings and classified with a graph neural network. Across an extensive sweep of different configurations per task on the SAND dataset, the sam...

[6] [6]

Generative AI Use Disclosure The authors take full responsibility for the accuracy, originality, and integrity of the final work and affirm that generative AI was used solely for language refinement and drafting assistance

[7] [7]

Self-supervised speech representation learning: A review,

A. Mohamed, H.-y. Lee, L. Borgholt, J. D. Havtorn, J. Edin, C. Igel, K. Kirchhoff, S.-W. Li, K. Livescu, L. Maaløe, T. N. Sainath, and S. Watanabe, “Self-supervised speech representation learning: A review,”arXiv preprint arXiv:2205.10643, 2022

arXiv 2022

[8] [8]

Amyotrophic lateral sclerosis,

M. C. Kiernan, S. Vucic, B. C. Cheah, M. R. Turner, A. Eisen, O. Hardiman, J. R. Burrell, and M. C. Zoing, “Amyotrophic lateral sclerosis,”The Lancet, vol. 377, no. 9769, pp. 942–955, 2011

2011