Sunday, Nov. 22, 2026 A Vik Media Group publication
Mongevity Longevity in Motion
Leer en español

Voice recordings added useful clues to dementia screening in a Korean study

Alzheimer's and dementiaPart of a story we are following · 14 articles · See every update
Men in paris park playing chess.
Men in paris park playing chess.Photo: zoetnet / Wikimedia Commons, CC BY 2.0

Your voice may one day be part of a memory check. Researchers found that the way people answer questions during a routine screening carries useful clues about dementia.

In a study of 1,091 people at five South Korean university hospitals, a computer model that combined voice recordings, drawings and clinical information sorted people with normal thinking from people with dementia with about 90% accuracy. The voice recordings did most of the work. It is early research, not a test you can get today, but the researchers say it may offer a practical, scalable way to add information to screening that already happens.

How the study worked

Researchers recorded people’s spoken answers and their pentagon drawings during a standard memory screening. The group included people with normal thinking, people with mild cognitive impairment, and people with dementia. A computer model learned to tell the groups apart from the voice, the drawings, and clinical details such as age and test scores. The work was published on November 11, 2026, in Geroscience. The model was tested by repeatedly splitting this one group of people into parts, and it had the hardest time telling normal thinking from mild impairment.

What is the MMSE?

The Mini-Mental State Examination, or MMSE, is a short test of memory and thinking. A clinician asks questions, and you answer out loud. One task is copying a drawing. The result is a score.

Usually, only that score is kept. In this study, the researchers also kept the audio of the answers and the drawing. They asked what those extras could add.

What they found

When the model had to sort people into three groups (normal thinking, mild cognitive impairment, dementia), it got about 68% right, or 67.9% to be exact. Most of its mistakes were between normal thinking and mild impairment. Those two are close together, and they are hard to tell apart.

The picture was much clearer when the question was simpler. For telling people with normal thinking from people with dementia, the combined model was right about 90% of the time, or 90.1% exactly.

Here is how each piece contributed:

  • Voice: models using voice did better than models using drawings alone.
  • Voice plus clinical data: this combination told the groups apart better than the clinical data by itself, including MMSE scores.
  • Drawings: the pentagon drawings added only a small benefit. Their main contribution was in spotting mild cognitive impairment.

In short, the recording of how someone answers held information that the score alone did not. The researchers say combining voice with MMSE scores may be a practical, scalable way to screen for dementia.

What you can do today

Nothing in your routine needs to change. Voice analysis like this is not something you can use at home, and it is not part of standard screening yet.

What you can do is keep a simple note. If you or someone you love has noticed changes in memory or thinking, write down what you have seen and when. Bring the list to your next checkup so your doctor can decide whether a memory screening makes sense.

For clinicians and researchers

Design: Multicenter development and validation study of a multimodal deep learning model in 1091 participants from five university hospitals in South Korea, classifying cognitively normal, mild cognitive impairment and dementia using voice, pentagon drawings, and demographic and clinical data; five-fold cross-validation.

Main result: Three-class multimodal accuracy was 67.9% (95% CI 65.9-70.0), AUC 0.84 (95% CI 0.82-0.85). Cognitively normal vs. dementia: accuracy 90.1% (95% CI 88.4-91.8), AUC 0.97 (95% CI 0.95-0.97). Voice outperformed drawings.

Main limitation: Validation was by five-fold cross-validation within the same cohort; the abstract reports no external validation, and misclassification was mainly between cognitively normal and mild cognitive impairment.

Open question: Whether voice-based models hold up in independent external cohorts and in routine screening settings, and whether they improve discrimination of mild cognitive impairment from normal cognition, remains to be tested.

Got a question? Ask us ›

Get the best of longevity science by emailThe most useful new studies, in plain language and with the source. Free.

We'll email you once to confirm. Occasional messages from our sponsors, clearly labeled. Unsubscribe anytime. We never ask for health data.
Cite this article
AMA
Mongevity Desk. Voice recordings added useful clues to dementia screening in a Korean study. Mongevity. Published November 22, 2026. Accessed November 22, 2026. https://mongevity.science/brain/2026/11/voice-recordings-dementia-screening/
Vancouver
Mongevity Desk. Voice recordings added useful clues to dementia screening in a Korean study [Internet]. Mongevity; 2026 Nov 22 cited 2026 Nov 22]. Available from: https://mongevity.science/brain/2026/11/voice-recordings-dementia-screening/
APA
Mongevity Desk. (2026, November 22). Voice recordings added useful clues to dementia screening in a Korean study. Mongevity. https://mongevity.science/brain/2026/11/voice-recordings-dementia-screening/
For your website
<a href="https://mongevity.science/brain/2026/11/voice-recordings-dementia-screening/">Voice recordings added useful clues to dementia screening in a Korean study</a>

Free to share or quote with credit and a link.

Questions and answers

Can a voice recording diagnose dementia?
No. In this study, a computer model used voice recordings along with other information to sort people into groups. It is research, not a diagnosis or a test available to you.
How accurate was the model?
For telling normal thinking from dementia, it was right about 90% of the time. For sorting three groups, including mild cognitive impairment, it was right about 68% of the time.
What is the MMSE?
The Mini-Mental State Examination is a short memory and thinking test given by a clinician. It produces a score.
Did the drawings help?
Only a little. The pentagon drawings added a small benefit, mostly for spotting mild cognitive impairment. Voice added more.

Sources

  1. Geroscience: What do voice and drawings recorded during the MMSE add to its score? A multicenter multimodal deep learning study of dementia screening (PMID 43009683 · DOI 10.1007/s11357-026-02647-4) · 2026-11-22

More on this story

Report an error