Abstract
Early detection of cognitive decline is crucial for enabling interventionsthat can slow neurodegenerative disease progression. Traditional diagnosticapproaches rely on labor-intensive clinical assessments, which are impracticalfor frequent monitoring. Our pilot study investigates voice assistant systems(VAS) as non-invasive tools for detecting cognitive decline throughlongitudinal analysis of speech patterns in voice commands. Over an 18-monthperiod, we collected voice commands from 35 older adults, with 15 participantsproviding daily at-home VAS interactions. To address the challenges ofanalyzing these short, unstructured and noisy commands, we propose Cog-TiPRO, aframework that combines (1) LLM-driven iterative prompt refinement forlinguistic feature extraction, (2) HuBERT-based acoustic feature extraction,and (3) transformer-based temporal modeling. Using iTransformer, our approachachieves 73.80% accuracy and 72.67% F1-score in detecting MCI, outperformingits baseline by 27.13%. Through our LLM approach, we identify linguisticfeatures that uniquely characterize everyday command usage patterns inindividuals experiencing cognitive decline.