Dysarthria speech models improve by tailoring to specific causes
Per-Aetiology Contrastive Severity Embeddings with Phonological Pseudo-Labelling for Multilingual Dysarthric Speech
Computation and LanguageSound
Summary
Understanding how severe speech problems are is important for helping people with conditions like cerebral palsy, Parkinson's disease, or ALS. This paper shows that speech models trained separately for each cause work better than one model trying to handle all causes together. The authors also use a new way to label speech without extra clinical input, which helps improve accuracy. The research tested these ideas across multiple languages and found better results when using tailored models and carefully chosen data.
What this means in practice
- •For speech technology developers: Develop specialized speech severity models tailored to each neurological cause to improve assessment accuracy across multiple languages.
- •For clinical speech pathologists: Use enhanced severity embeddings with pseudo-labelling to assist automatic monitoring of speech impairments from different neurological conditions.
Authors
Bernard Muller, Antonio Armando Ortiz Barrañón, LaVonne Roberts
Abstract
Most multilingual dysarthria-severity systems either train on a single aetiology-language pair or pool heterogeneous aetiologies into one label space. We test that pooling assumption with four matched HuBERT-base contrastive embedding models under a shared backbone, training recipe, corpus registry and held-out evaluation: one mixed-aetiology baseline and three aetiology-specific models for cerebral palsy (CP), Parkinson's disease (PD) and amyotrophic lateral sclerosis (ALS). Training combines clinically labelled speech with ordinal pseudo-labels from a training-free phonological profiling method [1], [2]. On speaker-disjoint, leakage-filtered held-out subsets, the per-aetiology models outperform the mixed baseline across all three target aetiologies: CP (macro F1 0.829 vs 0.676, +22.6 % relative), PD (0.715 vs 0.511, +40.0 %) and ALS (0.788 vs 0.596, +32.3 %). On CP, adding 144 SAP and 44 CDSD pseudo-labelled speakers lifts macro F1 from 0.786 to 0.829 over a clinical-only CP model (+4.3 percentage points). Training data span three to seven languages per aetiology. We position this as a controlled comparison of label-space design choices and discuss pseudo-label calibration, split hygiene, and confidence-thresholded deployment as important limitations for future work.