Mitigating Scoring Errors and Compensating for Nonverbal Subtests in Speech-Based Dementia Assessment

2026-06-17

AI Summary

Current speech-based assessments for dementia screening are limited by transcription errors and the omission of non-linguistic subtests—such as those evaluating motor skills—which compromises diagnostic accuracy. This study addresses these limitations in the context of the German standardized screening instrument “Syndrom-Kurz-Test” by introducing a novel approach that integrates Whisper-derived speech embeddings with post-transcription processing scores. Through multimodal feature fusion and regression modeling, the proposed method effectively corrects scoring bias and closely approximates expert global ratings, even when non-linguistic tasks are excluded. The framework significantly enhances the completeness and robustness of speech-based evaluation, demonstrating strong correlation between model predictions and clinician scores, accurate discrimination across cognitive status groups, and sustained high performance despite missing subtest data.

Abstract

Early detection of cognitive impairment relies on neuropsychological tests to minimize subjectivity by assessing multiple cognitive domains. Speech-based evaluation can support diagnostics and improve accessibility, but transcription errors and the omission of nonverbal subtests (e.g., motor skills) limit accuracy. Beyond conventional test scores, speech-derived features can provide additional insights into cognitive status. This study investigates the speech-based evaluation of the German "Syndrom-Kurz-Test," a standardized dementia screening test comprising verbal and motor subtests. We train models that integrate transcript-derived scores and Whisper embeddings per verbal subtest to reduce scoring errors. To compensate for missing motor subtests, we then leverage these fused representations to approximate expert overall ratings. Despite omitting subtests, our models strongly correlate with expert ratings and efficiently and accurately discriminate between cognitive status groups.

Problem

• speech-based dementia assessment
• scoring errors
• nonverbal subtests
• cognitive impairment
• neuropsychological tests

Innovation

• speech-based assessment
• Whisper embeddings
• cognitive impairment detection
• multimodal fusion
• dementia screening