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

The study presents a novel approach to enhance the accuracy of the German "Syndrom-Kurz-Test" for dementia screening by integrating transcript-derived scores with Whisper embeddings to mitigate scoring errors from verbal subtests. The models developed demonstrate a strong correlation with expert ratings, effectively compensating for the absence of nonverbal subtests while maintaining high discrimination accuracy between cognitive status groups. This advancement highlights the potential of speech-derived features in improving the reliability of cognitive assessments, which is crucial for practitioners in AI-driven diagnostic tools.