Bridging the Perceptual-Statistical Gap in Dysarthria Assessment: Why Machine Learning Still Falls Short

TL;DR

This paper analyzes why automated dysarthria assessment tools struggle to match human expert performance, introducing the perceptual-statistical gap as a core obstacle. It synthesizes insights from human perceptual processes, feature representations, and learning limits, highlighting how label noise and feature insufficiency cap ML performance. It then proposes concrete strategies—perceptually motivated features, self-supervised pretraining, ASR-informed objectives, multimodal fusion, and human-in-the-loop approaches—along with experimental protocols that align evaluation with clinical goals. The work emphasizes clinically meaningful evaluation and interpretability to bridge the gap between ML models and expert dysarthria assessment, aiming for robust, interpretable, and transferable tools.

Abstract

Automated dysarthria detection and severity assessment from speech have attracted significant research attention due to their potential clinical impact. Despite rapid progress in acoustic modeling and deep learning, models still fall short of human expert performance. This manuscript provides a comprehensive analysis of the reasons behind this gap, emphasizing a conceptual divergence we term the ``perceptual-statistical gap''. We detail human expert perceptual processes, survey machine learning representations and methods, review existing literature on feature sets and modeling strategies, and present a theoretical analysis of limits imposed by label noise and inter-rater variability. We further outline practical strategies to narrow the gap, perceptually motivated features, self-supervised pretraining, ASR-informed objectives, multimodal fusion, human-in-the-loop training, and explainability methods. Finally, we propose experimental protocols and evaluation metrics aligned with clinical goals to guide future research toward clinically reliable and interpretable dysarthria assessment tools.