00:00Short smartphone voice recordings carried clues to type 2 diabetes.
00:04ClickLab's researchers report,
00:05The study used voice data alone, with no other health details added.
00:10Researchers analyzed 14,159 voice recordings from 461 adults.
00:16Each person read short English sentences into a phone.
00:19An AI model turned the audio into spectrograms, visual maps of sound frequencies.
00:24It then searched those images for patterns linked to diabetes.
00:28Balanced accuracy reached 71.3% in women and 72.4% in men.
00:34Its AUC, a standard accuracy score, was 0.788 for women and 0.779 for men.
00:41The findings appear in the IEE Journal of Biomedical and Health Informatics.
00:46We can literally see many more clues, said lead investigator Jan Fossat of Click Applied Sciences.
00:52But these scores show moderate accuracy, not a diagnosis.
00:55The study also did not describe testing on a separate group of patients.
01:00For now, voice clues to diabetes remain an early research finding.
01:04Disclosure.
01:05This video contains stock footage and content created or enhanced using AI-assisted tools.
