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

2026-07-28

AI Summary

This study addresses the challenge of modeling multiple speech utterances in relation to disease severity for amyotrophic lateral sclerosis (ALS) biomarker analysis. The authors propose a novel subject-level multi-utterance graph learning framework that integrates self-supervised speech embeddings with graph neural networks. Specifically, utterance-level embeddings are extracted using wav2vec 2.0 or HuBERT, from which k-nearest neighbor graphs are constructed and subsequently processed via graph convolutional networks (GCN), graph attention networks (GAT), or graph isomorphism networks (GIN) for aggregation and prediction. Evaluated on the SAND dataset, the HuBERT+GIN configuration achieves macro-F1 scores of 0.73 for ALS detection and 0.69 for progression prediction, significantly outperforming existing baselines and demonstrating the approach’s effectiveness and innovation under low-resource conditions.

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.

Problem

• ALS detection
• dysarthria severity
• disease progression prediction
• speech biomarkers
• acoustic analysis

Innovation

• graph neural networks
• self-supervised speech embeddings
• multi-phonation modeling
• ALS progression prediction
• pretrained audio representations