![[Adobe Stock]](https://sp-ao.shortpixel.ai/client/to_webp,q_glossy,ret_img,w_750,h_375/https://www.psypost.org/wp-content/uploads/2025/07/brain-scans-1-750x375.jpg)
An automated analysis of short voice recordings can produce an estimate of “speech age” that runs older in people with dementia than in cognitively healthy adults, according to a study of nearly 3,000 adults across Latin America published in Science Advances. The gap between this speech-based estimate and a person’s actual age was also linked to brain-scan and DNA-based measures of aging, a blood marker of Alzheimer’s disease, and lifetime social hardship.
To estimate whether a person’s body appears older or younger than their calendar age, scientists use tools known as biological aging clocks. A biological aging clock is a statistical model that analyzes physical or behavioral data to predict a person’s age. The difference between this predicted age and the person’s actual chronological age is called an age gap, where a positive gap is typically interpreted as accelerated aging and a negative gap as healthier, delayed aging.
Most established aging clocks rely on magnetic resonance imaging (MRI) to measure brain tissue volume and activity, or blood draws to examine chemical modifications to DNA known as epigenetic methylation. While a study covered by PsyPost in early 2026 showed that blood tests can detect dementia-linked proteins in Latin American populations, both neuroimaging and molecular blood analyses remain expensive and logistically difficult to deploy in lower-income regions. Behavioral age clocks based on questionnaires and cognitive exams also require trained specialists and can be skewed by a patient’s literacy level.
Natural speech may offer an accessible, non-invasive alternative for tracking how the brain and body age. Speaking requires the rapid coordination of multiple cognitive and physical systems, which shift during normal aging and break down more rapidly in neurodegenerative diseases. A slower speaking rate and longer pauses between words or syllables are thought to reflect declines in mental processing speed and word retrieval.
Changes in vocal pitch and the emotional tone of words can capture the flattened emotional expression often seen in dementia. Shifts in vocabulary, such as relying on words for physical objects rather than abstract ideas or using less precise terms, point to the erosion of semantic memory, which is the brain’s system for storing general knowledge and facts.
The research was led by Hernan Hernandez of the Latin American Brain Health Institute at Universidad Adolfo Ibáñez in Chile, with senior authors Agustin Ibanez and Adolfo M. García. Ibanez is a professor of global brain health at the Global Brain Health Institute at Trinity College Dublin, scientific director of the Latin American Brain Health Institute, and strategic scientific director at the Barcelonaβeta Brain Research Center. García directs the Cognitive Neuroscience Center at Universidad de San Andrés in Argentina and is affiliated with the Global Brain Health Institute at the University of California, San Francisco, and the Universidad de Santiago de Chile.
They sought to test whether discrepancies between a person’s chronological age and their speech-predicted age could distinguish between different types of dementia and mirror internal biological aging.
“This study grew from two related goals,” Ibanez told PsyPost. “First, our group has been developing multiple aging clocks to capture the cumulative burden of biological, brain, behavioral, and exposome processes across aging and dementia. Second, many established biomarkers are expensive and difficult to scale, particularly in low-resource settings. We therefore wanted to test whether natural speech could provide a low-cost, non-invasive measure that captures several dimensions of aging simultaneously.”
“We wanted to establish whether speech recordings (an affordable, naturalistic, scalable data source) could offer reliable predictions of aging, considering multiple sociobiological factors in both healthy older adults and individuals with dementia,” García added.
The researchers analyzed data from 2,928 native Spanish speakers across five Latin American countries: Colombia, Peru, Mexico, Chile, and Argentina. The participants were recruited through a multi-country research network and had an average age of about 65 years, with women making up 67 percent of the sample. The cohort included 1,504 cognitively healthy individuals and 1,424 patients diagnosed with various neurodegenerative conditions.
Among the patient group, 24 had mild cognitive impairment (MCI), a transitional stage between normal aging and dementia, and 1,068 had Alzheimer’s disease (AD), the most common form of dementia characterized by progressive memory loss. Another 332 patients had frontotemporal dementia, a group of disorders caused by damage to the brain’s frontal and temporal lobes. The team divided the frontotemporal dementia patients into 255 individuals with variants that primarily affect behavior and movement, and 77 individuals with variants that primarily impair language production and comprehension.
Every participant completed seven brief speaking tasks in a quiet room, including describing an animated video, retelling a short story immediately and after a 20- to 30-minute delay, and naming as many words as possible starting with a specific letter or belonging to a category like animals within 60 seconds. The researchers processed the audio recordings using TELL, an automated speech-analysis app for which García serves as founder and chief scientific officer. Its pipeline removed background noise, isolated the participant’s voice from the examiner’s, and generated text transcripts.
From these recordings and transcripts, the team extracted over 700 distinct acoustic and language measurements and grouped them into seven composite categories: speech timing, vocal pitch, emotional content, word frequency, verbosity, semantic concreteness, and semantic granularity (how specific a word is, such as saying “robin” instead of “bird”).
The scientists trained a machine learning model on these seven speech categories to predict chronological age, and then subtracted each person’s actual age from their predicted age to calculate their speech age gap. On average, the model’s age estimates missed participants’ actual ages by about nine years and accounted for roughly 44 percent of the variation in age across the sample; after a statistical correction commonly applied to age-prediction models, predicted and actual ages correlated at 0.74. Verbosity and speech timing were the most influential features indicating preserved or younger-sounding speech, while vocal pitch variations and negative emotional word content were the strongest indicators of accelerated speech aging.
When comparing the diagnostic groups, cognitively healthy adults had the lowest speech age gaps, whereas every patient group showed larger gaps, reflecting older-sounding speech. Patients with language-dominant frontotemporal dementia displayed the largest speech age gaps, followed by those with behavioral or motor frontotemporal dementia, and then those with Alzheimer’s disease and mild cognitive impairment.
These group differences held up in sensitivity analyses that controlled for chronological age, sex, years of education, country of origin, medical comorbidities, and the number of languages a participant spoke. The researchers also examined whether speaking multiple languages was linked to lower speech age gaps in a subset of participants, finding a slight trend toward younger-sounding speech among multilingual speakers that did not reach statistical significance.
In additional comparisons, the full speech clock separated the diagnostic groups more effectively than any isolated acoustic or language category on its own. “Aging leaves a distributed signature across multiple dimensions of speech—timing, lexical choice, semantic properties, prosody, and verbal output,” García explained. “No single feature tells the whole story; the informative signal emerges from their combination.”
“All too often, brain disorders are conceived as primarily affecting or broadly sparing language,” García said. “This usually reflects a poor operationalization of language. When rich characterizations of speech and discourse are used, ensuing signals can improve the detection, differentiation, and characterization of disorders which are traditionally not associated with language anomalies as a defining feature.”
The researchers next examined how these speech age gaps related to cognitive test scores, blood biomarkers, and lifetime social environment. Across the cohort, larger speech age gaps were associated with worse performance on clinical tests of memory, executive function, everyday functional ability, and both verbal and non-verbal tasks. The link was stronger for tests with a verbal component, such as retelling a story, than for tests the authors classified as nonverbal, such as copying and later redrawing a complex figure. In a subset of participants who underwent blood testing, higher speech age gaps correlated with elevated blood plasma levels of phosphorylated tau-217, a protein that accumulates in the brain during Alzheimer’s disease, specifically among the Alzheimer’s patients.
The team also evaluated each participant’s social exposome, a comprehensive measure combining 319 survey items covering lifetime socioeconomic factors such as education, food insecurity, financial stress, healthcare access, and childhood adversity. Higher speech age gaps were linked to a more adverse social exposome among healthy controls and Alzheimer’s patients, suggesting that cumulative social hardship may be reflected in how old a person sounds. This aligns with earlier research covered by PsyPost in 2025, conducted by several members of the same research network, which found that a heavier lifetime burden of social adversity was associated with poorer cognition and disrupted brain connectivity in Latin American cohorts.
“I was particularly struck that speech reflected not only brain and molecular aging but also the social exposome,” Ibanez said. “It suggests that the cumulative environments people live in (their socioeconomic circumstances, education, access to resources, and other life-course exposures) may leave measurable traces in something as everyday as speech.”
To verify whether the speech clock reflected internal biological aging, the authors compared the speech age gaps against established brain and epigenetic clocks in subsets of participants with available neuroimaging and blood DNA data. The speech age gaps showed moderate-to-strong positive correlations with brain age gaps derived from structural MRI scans measuring gray matter volume (correlation of 0.42), functional MRI scans measuring how brain regions communicate at rest (correlation of 0.49), and a combined brain clock (correlation of 0.53) across healthy controls and all dementia groups. In addition, larger speech age gaps were positively associated with accelerated biological aging across three independent epigenetic clocks that measure DNA methylation patterns in white blood cells, particularly among healthy adults and patients with Alzheimer’s disease, though these links were weaker (correlations of about 0.2) than those with the brain clocks.
When the team examined specific brain networks, they found that the speech clock tracked whole-brain aging more closely than the aging of isolated brain circuits, including the brain’s dedicated language network. Only the language network and the salience network, which helps the brain monitor speech and allocate attention, showed small links to the speech clock, while other individual brain networks showed no statistically significant relationship.
“One striking finding was how strongly the speech clock aligned with brain-age measures across several dementia syndromes, despite being derived solely from speech,” García told PsyPost. “We were also intrigued that whole-brain aging measures related more strongly to speech aging than most individual functional networks did. That suggests speech may reflect the coordinated state of distributed brain systems rather than the condition of a single language circuit.”
“Speech appears to contain much more information about aging than we usually assume,” Ibanez noted. “Our speech-derived aging measure was associated not only with cognitive and clinical impairment, but also with brain aging, epigenetic aging, p-tau217, and social exposures. This suggests that a few minutes of speech may eventually provide an accessible window into broader brain and biological health.”
One independent researcher’s assessment highlighted both the scale of the multi-country effort and the nuances of interpreting speech-based aging markers across neurodegenerative conditions. The cohort’s language-dominant frontotemporal dementia group included patients with non-fluent and semantic variants of primary progressive aphasia, which are neurological syndromes that gradually erode a person’s ability to speak or understand word meanings.
“Hernandez et al. have derived a novel marker of aging by analyzing acoustic and linguistic features of speech samples using a machine-learning algorithm to predict age and by subsequently computing deviations of predicted age from actual age,” Mai-Carmen Requena-Komuro, a postdoctoral researcher at the Institute of Systems Neuroscience at the University Medical Center Hamburg-Eppendorf in Germany who was not involved in the study, told PsyPost. “They show that these ‘speech age gaps’ (SAGs) enable us to distinguish healthy elderly people from people diagnosed with dementia in a large-scale Spanish-speaking cohort living across several countries in Latin America.”
“Speech deficits are known characteristics of language variants of frontotemporal dementia (non-fluent variant primary progressive aphasia and semantic variant primary progressive aphasia, both included in the current cohort), although the specific speech domains targeted differ,” Requena-Komuro said. “Speech deficits have recently caught more attention in patients diagnosed with Alzheimer’s disease (there is in fact a language variant of AD known as logopenic primary progressive aphasia).”
As with all research, there are some caveats to consider, including how participants were screened. To qualify as cognitively healthy in the study, participants had to demonstrate functional independence and score above 25 out of 30 on the Mini-Mental State Examination (MMSE), a standard 30-point questionnaire used to screen for cognitive impairment. In addition, the sample sizes for certain subgroups, particularly the 24 patients with mild cognitive impairment and the 77 patients with language-dominant frontotemporal dementia, were relatively small, and not all participants had blood, brain scan, or epigenetic data available. Because the study relied on cross-sectional data collected at a single point in time, it cannot establish whether an older-sounding voice actually predicts future cognitive decline or tracks the speed of aging over the years.
“I do find it admirable that the authors managed to recruit such a diverse and large cohort of people living with dementia across so many countries,” Requena-Komuro noted. “However, there are a couple of points that made me slightly skeptical: (i) the cut-off score of the MMSE for the healthy controls is set at 25, but the questions are relatively easy so I would not expect elderly adults to score anything below 28 (the maximum is 30). Some of these healthy controls may therefore have belonged to the MCI group instead; (ii) it is unclear whether the regression models aimed at predicting the age of participants included both healthy elderly and elderly with dementia; I would expect an initial model focusing on healthy elderly only to then better understand where and by how much elderly with dementia deviate.”
“Not all speech features go in the same direction, yet they all contribute to the ‘speech age gap’ (SAG),” Requena-Komuro added. “I would therefore caution readers not to overinterpret the SAG as an overall marker of degraded speech. Overall, this study is rich in correlation plots, but correlation does not mean causation. The fact that the SAG correlates with many different factors does not provide much more explanation than these individual factors considered individually.”
Another consideration involves how different neurodegenerative conditions progress across the lifespan and how much speech patterns vary from person to person within the same diagnostic category. Frontotemporal dementia often begins at an earlier age than typical Alzheimer’s disease.
“Generally, a more hypothesis-driven study would have been more insightful,” Requena-Komuro explained. “Frontotemporal dementia (FTD) is an early-onset dementia; it is therefore not surprising that the SAG scores for these subgroups are significantly higher than that of the AD group (if the signs of dementia start earlier for FTD patients, when we compare them at the same age of AD patients, their symptoms are likely to be more severe). Perhaps this is one way of validating the SAG, but given the substantial variability in individual trajectories even within the same clinical group, the SAG could not be used as a standalone measurement to establish diagnosis or characterize disease progression. Unlike what the authors claim, I would therefore say that we are not ready to replace standard cognitive and neuropsychological tests for dementia.”
The study’s authors also describe the speech clock as a complement to existing evaluations rather than a stand-alone diagnostic test. “The speech clock captured clear differences between groups and showed especially strong links with brain-age measures, but it is best viewed as a signal that becomes informative when combined with other evidence, not as a stand-alone test of a person’s health or prognosis,” García said. “A positive speech-age gap should not be interpreted as a diagnosis or as proof that someone is literally ‘aging faster.'”
Because the study was conducted entirely among Spanish speakers in five Latin American nations using structured speaking tasks, the speech clock cannot be applied directly to other languages or casual everyday conversations without further testing. Different languages have unique grammatical rules and acoustic rhythms, meaning that the specific features that signal Alzheimer’s or frontotemporal dementia in Spanish may behave differently in other languages, particularly word-based features, which have carried over less well than timing features in earlier cross-language work. Future longitudinal studies tracking diverse, multilingual populations over time will be needed to determine whether automated voice analysis can serve as a routine screening tool in clinics and public health programs.
“Some associations were moderate rather than large, so this should be viewed as a complementary biomarker rather than a replacement for established clinical tools,” Ibanez said. “Aging clocks are estimations, not diagnoses or deterministic measures of how fast a specific person is aging. Importantly, this study was primarily cross-sectional, so we cannot yet infer individual aging trajectories or say that someone whose speech appears ‘older’ is necessarily aging faster over time.”
“A valuable next step is to determine whether speech-age gaps predict longitudinal cognitive and clinical change, especially across different languages,” García added. “Prediction of future states is a fundamental but rarely hit milestone in speech biomarker research.”
Moving forward, the team plans to explore how lifestyle and clinical interventions might influence vocal aging markers and how these tools can expand dementia research in regions with limited medical infrastructure.
“We also want to test whether interventions that improve brain health shift speech aging toward a healthier trajectory, and to validate these models across languages, cultures, and more naturalistic speech,” Ibanez said. “One important aspect of the study is that it included nearly 3,000 participants across five Latin American countries, which are populations that remain underrepresented in dementia research. Speech-based approaches could be particularly valuable in settings where MRI, PET, or sophisticated molecular biomarkers are difficult to access at scale.”
The study, “Speech clocks decode dementia phenotypes, social exposome, and biological aging,” was authored by Hernan Hernandez, Lizeth Katherine Pedraza, Hernando Santamaria-Garcia, Sebastian Moguilner, Agustina Legaz, Pavel Prado, Jhosmary Cuadros, Lucía Amoruso, Liset Gonzalez, Damián Dellavale, Juan Pablo Espinoza–Puelles, Javier Palma Espinosa, Cecilia Jarne, Fabio Mattiussi, Matías Caccia, Alejandro Sosa Welford, Nicolás Pelella, Jeremías Inchauspe, Franco J. Ferrante, Gonzalo Pérez, Marcelo Adrián Maito, Guido Rocatti, Maria Eugenia Godoy, Joaquin Migeot, Paulina Orellana, Ariel Caviedes, Martin Bruno, Leonel Takada, Andrea Slachevsky, Maria I Behrens, Bárbara Bruna, David Aguillon, Lina Zapata, Jose Alberto Avila-Funes, Nilton Custodio, Bruce Miller, Maria Luisa Gorno-Tempini, Stefanie Pina Escudero, Pablo Reyes, Kun Hu, Maira Okada de Oliveira, Carlos Coronel-Oliveros, Josephine Cruzat, Juan Felipe Cardona, Michael Corley, Irene B. Meier, Vaibhav A. Narayan, Enzo Tagliazucchi, Sandra Baez, Claudia Duran-Aniotz, Adolfo M. García, and Agustin Ibanez.