Can a One-Minute Speech Test Rival Traditional Tools for Detecting Alzheimer’s?

In this perspective, Atlantic Fellow Adolfo García and Iván Caro examine how a brief speech test could help make Alzheimer’s screening faster, more accessible, and more equitable around the world.

Can Speech Detect Alzheimer’s as Well as Traditional Tools?

Alzheimer’s disease affects more than memory. It also involves changes in how people speak, influencing pauses, word choice, and vocabulary complexity. Researchers can now capture these subtle patterns with digital tools, raising the possibility of faster and more equitable brain health assessments. Speech can be collected in seconds, analyzed automatically, and potentially deployed anywhere with a smartphone.

Now, enthusiasm must be matched with caution. Speech-based tools could make cognitive screening more accessible, especially in regions with limited access to specialists or advanced imaging. But before these approaches can be widely adopted, they must be shown to perform as well as established methods. This motivated our study’s simple but important question: can speech data match traditional tools in identifying Alzheimer’s disease?

Putting Speech to the Test

We evaluated 66 Spanish‑speaking participants from Latin America: 33 people with Alzheimer’s disease and 33 healthy older adults. Each participant completed two short verbal fluency tasks commonly used in cognitive assessments. In one task, they named as many animals as possible within one minute. In another, they generated words beginning with a specific letter. Though simple, these tasks contain a surprising amount of information about how the brain retrieves language.

Using our TELL app, we automatically analyzed two types of digital speech markers. The first captured timing, including pauses, speaking speed, and how speech unfolds over time. The second focused on the words themselves, including their length, frequency of use, and level of specificity.

To understand how meaningful these speech signals were, we compared them with several well‑established approaches to identify Alzheimer’s disease. These included memory and thinking tests, brain imaging that measures changes in brain structure, and measures of how different brain regions communicate with one another.

Machine‑learning models were trained separately on each type of data to determine how well they could distinguish people with Alzheimer’s disease from healthy participants. Would these highly scalable speech tests measure up to well-established cognitive and neuroimaging benchmarks?

Speech Held Its Ground

Speech‑based models were able to identify Alzheimer’s disease with performance that closely matched several traditional diagnostic approaches. Their performance closely matched approaches based on episodic memory testing, thinking skills, and structural brain imaging. In practical terms, this means that a brief speech task captured nearly as much diagnostic information as tools that often require longer testing sessions or specialized equipment.

Compared with episodic memory measures and MRI‑based brain markers—the strongest benchmarks in our study—speech showed only minimal drops in performance. At the same time, speech outperformed models based on executive functions and functional brain connectivity, which were less effective at separating patients from healthy individuals in this dataset.

Looking closely at the speech itself revealed clear linguistic patterns. Compared with healthy participants, people with Alzheimer’s tended to use shorter words and rely more heavily on very common vocabulary. They also tended to use more general words rather than highly specific ones, and their speech contained longer pauses while searching for words.

These patterns reflect disruptions in semantic memory, the system that stores knowledge about words and concepts. Our findings indicate that, when retrieving words becomes more difficult, people tend to rely on simpler, more accessible vocabulary. Interestingly, some of these speech markers also tracked changes in brain structure. For example, heavier reliance on common or shorter words was associated with reduced volume in frontal brain regions involved in controlled word retrieval.

Why This Matters for Global Brain Health

Traditional diagnostic tools for Alzheimer’s disease (such as MRI scans, specialized cognitive tests, or biomarker tests) are powerful but often require trained professionals, specialized equipment, and significant clinical infrastructure. Speech‑based approaches offer something different. Because verbal fluency tasks take only about a minute and can be analyzed automatically, they could enable scalable cognitive screening across clinics, communities, and remote settings. For regions such as Latin America and Africa, where dementia prevalence is rising but diagnostic resources remain limited, this accessibility could make a meaningful difference.

Our findings suggest that speech biomarkers are not a second-rate fallback to standard tools. As digital health technologies advance in Alzheimer’s research, the words we speak can increase equity without sacrificing precision.

Reference

Caro, I., Pérez, G., Valdés Bize, J., Ponferrada, J., Ferrante, F. J., Sosa Welford, A., Gauder, L., Olavarría, L., Henríquez, F., Ramos, T., Besnier, C., Ferrer, L., Gorno-Tempini, M. L., Slachevsky, A., Ibañez, A. & García, A. M. (2026). Benchmarking speech biomarkers of Alzheimer’s against cognitive and neural measures. Alzheimer's & Dementia 22, e71365.