Adapting Self-Supervised Speech Representations for Cross-lingual Dysarthria Detection in Parkinson's Disease
2026-03-23 • Computation and Language
Computation and LanguageSound
AI summaryⓘ
The authors address the challenge of detecting speech problems called dysarthria across different languages, which is hard because speech features often carry language-specific details. They introduce a method called language shift (LS) that adjusts speech representations from one language to better match another using examples from healthy speakers. Tested on speech data from people with Parkinson's disease in Czech, German, and Spanish, their method improved detection accuracy, especially when working across different languages. Their analysis suggests that LS helps by reducing language-specific clues in the speech data.
dysarthriacross-lingual detectionself-supervised speech representationslanguage shiftvector adaptationParkinson's disease speechoral DDK recordingsembedding spacemultilingual settings
Authors
Abner Hernandez, Eunjung Yeo, Kwanghee Choi, Chin-Jou Li, Zhengjun Yue, Rohan Kumar Das, Jan Rusz, Mathew Magimai Doss, Juan Rafael Orozco-Arroyave, Tomás Arias-Vergara, Andreas Maier, Elmar Nöth, David R. Mortensen, David Harwath, Paula Andrea Perez-Toro
Abstract
The limited availability of dysarthric speech data makes cross-lingual detection an important but challenging problem. A key difficulty is that speech representations often encode language-dependent structure that can confound dysarthria detection. We propose a representation-level language shift (LS) that aligns source-language self-supervised speech representations with the target-language distribution using centroid-based vector adaptation estimated from healthy-control speech. We evaluate the approach on oral DDK recordings from Parkinson's disease speech datasets in Czech, German, and Spanish under both cross-lingual and multilingual settings. LS substantially improves sensitivity and F1 in cross-lingual settings, while yielding smaller but consistent gains in multilingual settings. Representation analysis further shows that LS reduces language identity in the embedding space, supporting the interpretation that LS removes language-dependent structure.