Towards Dys-XAI: Influence-Based Explanations for Dysarthria Severity Assessment
For clinicians and speech therapists, this provides auditable, case-based explanations for AI-assisted dysarthria assessment, addressing a key barrier to clinical adoption.
The paper tackles the lack of interpretability in deep learning models for dysarthria severity assessment. It proposes an influence-based explainability framework that links predictions to specific training samples, validated by controlled deletion experiments showing systematic prediction shifts.
Dysarthria severity assessment is essential for therapy planning and longitudinal monitoring, yet manual perceptual rating is time-consuming and variable across clinicians. Although deep learning models achieve strong performance, their black-box nature limits clinical adoption. Existing speech explainability methods typically provide acoustic feature importance scores that are difficult for end-users to interpret. We propose an influence-based, instance-level explainability framework that explains each decision through supportive and competing training samples. Using gradient-based influence approximations, we compute per-utterance influence scores to identify supportive and competing training samples for each prediction. Controlled deletion experiments from 5 to 20 percent validate the explanations, showing that removing highly influential samples systematically shifts predictions. This approach provides auditable explanations by linking decisions to perceptible reference cases.