AI can now diagnose the severity of a speech disorder in five different languages without ever being taught those languages.
By analyzing how phonological features degrade in 'frozen' speech models, researchers found a universal biological signature of impairment. This allows for clinical-grade assessments in any language without the need for expensive, human-labeled training data.
Training-Free Cross-Lingual Dysarthria Severity Assessment via Phonological Subspace Analysis in Self-Supervised Speech Representations
arXiv · 2604.10123
Dysarthric speech severity assessment typically requires trained clinicians or supervised models built from labelled pathological speech, limiting scalability across languages and clinical settings. We present a training-free method that quantifies dysarthria severity by measuring degradation in phonological feature subspaces within frozen HuBERT representations. No supervised severity model is trained; feature directions are estimated from healthy control speech using a pretrained forced aligne