Biomarkers.

Area of Science:

• Neurolinguistics and Computational Linguistics
• Biomedical Engineering and Health Informatics

Background:

• The Toolkit to Examine Lifelike Language (TELL) is a web application developed to collect voice samples and extract speech biomarkers for dementia detection.
• The initial version of TELL successfully supported clinical evaluations globally, including in under-resourced regions.

Purpose of the Study:

• To introduce TELL 2.0, an upgraded version of the TELL application with advanced features for dementia evaluation.
• To enhance data collection, processing, analysis, and visualization capabilities for speech biomarker research.

Main Methods:

• TELL 2.0 incorporates videocall-based and offline data collection, supporting lossless audio and transcriptions via a local Whisper model.
• An audio standardization pipeline (normalization, noise reduction, harmonization) and new interpretable metrics for lexical, motor, semantic, and self-referential language were added.
• A machine learning-based prediction tool compares individual results to dementia profiles (Alzheimer's, MCI, bvFTD) and visualizes findings against normative data.

Main Results:

• Consistent audio and transcription quality were maintained across different data collection modes.
• The new speech biomarkers accurately differentiate individuals with specific neurological disorders from healthy controls, with accuracies ranging from 85% to 93%.
• The updated visualization interface offers clinicians actionable insights, aligning metrics with established neuropsychological frameworks.

Conclusions:

• TELL 2.0 expands the reach and effectiveness of speech biomarkers for dementia assessment, building upon the success of its predecessor.
• The project emphasizes translating advanced research into practical, scalable, and cost-effective dementia evaluation tools.