TELL: A web‐based app to capture speech and language markers of neurodegeneration

Abstract

AbstractBackgroundOver the last decade, the search for scalable markers of neurodegeneration has been boosted by automated speech and language analysis (ASLA). Participants are asked to produce natural discourse, generating acoustic and linguistic data that can be digitally extracted and analyzed with statistical or machine learning methods. This approach has revealed subtle patterns that differentiate among disorders, predict core symptom severity, correlate with anatomo‐functional brain alterations, and identify autopsy‐confirmed pathology years before death. Accordingly, multiple studies have described ASLA as a promising resource for patient assessment. However, this translational potential remains unfulfilled, as the underlying technologies are unavailable for immediate clinical use. Aiming to tackle this issue, we developed the Toolkit to Examine Lifelike Language (TELL), a web‐based app for collecting, storing, analyzing, and visualizing speech data.MethodTELL version 1.0 was developed by an interdisciplinary team of language, brain health, and data science experts. We designed a speech elicitation protocol comprised of validated spontaneous, semi‐spontaneous, and non‐spontaneous tasks with different cognitive demands. Each task is analyzed via ASLA algorithms capturing word frequency, affective valence, and emotional magnitude, as well as rhythm‐ and pitch‐related features. On‐the‐fly results are offered through intuitive visualizations for each metric. The measures are complemented with a language profile survey. Audio files, transcriptions, and data matrices can be downloaded for offline analysis. The app runs on multiple devices, requires no installation, and offers multi‐language capabilities.ResultsTELL version 1.0 is now being used in 10 sites worldwide, mainly including low‐income countries. Preliminary results from persons with Alzheimer’s and Parkinson’s disease, as well as frontotemporal dementia, show that its metrics are sensitive for disease identification, discrimination, and monitoring. Updates have been implemented drawing from users’ feedback. Funds have been secured to develop TELL version 2.0, with novel metrics, improved speech‐to‐text algorithms, and a streamlined front end.ConclusionTELL turns basic language science methods into clinically usable tools, offering an objective, low‐cost, non‐invasive complement to classical assessments. More generally, it illustrates how digital technologies can foster equitable insights on brain health. Ongoing updates will further increase its capacity to establish scalable markers of diverse neurodegenerative diseases.