Particle.news

AI Speech Tool Flags Risk of Type 2 Diabetes

Phone- and app-based screening could widen reach pending clinical trials and equity safeguards.

Overview

  • Researchers reported on Monday that their speech-analysis model gave higher risk scores to people who self-reported type 2 diabetes in 80% of 7,319 UK participants.
  • The system was trained on more than 63,000 voice recordings from over 21,000 people in the UK and US and matched home blood-test results in 75% of a 801-person subgroup.
  • The tool detects vocal signs linked to diabetes, such as hoarseness, breath-control changes and vocal strain that can arise from nerve damage and reflux, which can affect the voice box and airflow.
  • Performance was lower for Black participants and for people with obesity, high blood pressure or heart disease, which researchers say likely reflects underrepresentation and raises equity and bias concerns.
  • Authors and Diabetes UK stress the model is a screening aid not a diagnostic test and call for clinical-setting trials, transparent subgroup reporting, and stronger safeguards before any wider use.