Could speech offer a window into aging?

By Dr. Liji Thomas, MDReviewed by Lauren HardakerOct 7 2026

A speech clock built from nearly 3,000 people tracked dementia, cognition, social adversity, and biological aging, raising the prospect of a low-cost window into how we age.

Study: Speech clocks decode dementia phenotypes, social exposome, and biological aging. Image credit: adriaticfoto/shutterstock.com

A recent study published in the journal Science Advances presents a large-scale speech clock that captured variation across multiple levels of aging, making it a potential biomarker for aging research, including in underrepresented areas worldwide.

Could speech provide a simpler measure of aging?

Aging clocks are designed to capture a simple but important idea: the body and brain do not necessarily age at the same pace as the calendar.

Rather than relying on chronological age alone, researchers can estimate biological aging in several ways. Biobehavioral age gaps (BBAGs), for example, can capture aging differences associated with social and behavioral exposures, while brain age gaps (BAGs) measure the difference between a person's estimated brain age and their chronological age. Epigenetic clocks take another approach, using patterns of DNA methylation, while multidimensional clocks bring different measures together to capture aging across multiple biological and behavioral systems.

But many of these approaches depend on costly tests, specialized equipment, or complex data collection, limiting their potential for large-scale use, particularly in lower-resource settings. The researchers therefore highlight the need for “objective, automated, affordable, noninvasive aging markers enabling timely, large-scale, equitable deployment in low-income, underrepresented regions with high dementia burden.”

Speech could offer one such window into aging. How people speak changes with both healthy aging and dementia as cognition and motor control are affected. Slower processing can show up as changes in speech timing, including longer pauses and less efficient word retrieval. Affect may also become flatter with age, and more markedly in dementia, with changes evident in pitch-related and emotional features. Semantic aspects of language also change: healthy aging can include vocabulary enrichment, whereas dementia can involve deterioration of semantic knowledge.

These changes can be captured automatically from speech recordings and transcripts, potentially offering a comparatively inexpensive and scalable source of aging information. By analyzing acoustic and linguistic features, researchers can estimate a person's “speech age” and compare it with their chronological age to calculate a speech age gap (SAG).

The researchers set out to determine whether these speech age gaps were more than simply a measure of how old someone sounded. They examined whether SAGs aligned with different dementia phenotypes and with cognitive, social, neuroimaging, molecular, and epigenetic measures of aging.

Nearly 3,000 participants put the speech clock to test

The study included 2,928 Spanish-speaking participants across multiple countries (Colombia, Chile, Argentina, Mexico, and Peru), who were part of the Multi-Partner Consortium to Expand Dementia Research in Latin America (ReDLat). It includes a wide spectrum of cognitive performance: cognitively healthy controls (HCs), persons with mild cognitive impairment (MCI), Alzheimer’s disease (AD), non-language-dominant frontotemporal dementia (nldFTD), and language-dominant frontotemporal dementia (ldFTD). The cohort comprised 1,504 HCs, 24 people with MCI, 1,068 with AD, 255 with nldFTD, and 77 with ldFTD.

Speech recordings were taken from all participants using standardized Spanish-language tasks, including video description, verbal fluency, and immediate and delayed story-retelling tasks. Subsets of participants also had available plasma biomarkers (phosphorylated tau217, p-Tau217), clinical and cognitive scores converted to a composite measure, and social exposome measures. Using an automated pipeline, SAGs were calculated, with SAG>0 indicating an older-appearing speech profile and SAG<0 a better-preserved or younger-appearing one.

These cross-sectional gaps indicate how old or young a person's speech appeared relative to their chronological age, rather than demonstrating that they were actually aging faster or slower over time.

Composite speech features provided an aging signal that could be used to predict SAGs.

Speech age gaps widen across dementia groups

The speech clock revealed clear differences between healthy controls and patient groups. SAGs were largest in people with language-dominant frontotemporal dementia (ldFTD), followed by non-language-dominant FTD (nldFTD), Alzheimer’s disease (AD), and mild cognitive impairment (MCI).

These gaps reflect how old a person’s speech appeared relative to their chronological age.

Older-appearing speech tracks cognition and tau

Larger SAGs were also linked to poorer cognitive and clinical performance, including memory, executive function, and global cognition. The relationship with memory was particularly strong in AD and nldFTD, while associations were generally stronger for linguistic than nonlinguistic abilities.

Larger SAGs also accompanied higher plasma p-Tau217 levels overall. When diagnostic groups were examined separately, however, this association was significant only in AD.

Social adversity leaves a signal in speech

The speech clock also aligned with the social exposome, which captures lifelong influences including education, food insecurity, financial circumstances, healthcare access, and early-life experiences.

Across the cohort, larger SAGs were associated with more adverse social conditions, with significant within-group relationships seen in healthy controls and people with AD. The findings show an association, not evidence that social adversity caused older-appearing speech.

Speech aging aligns with biological clocks

Older-appearing speech also tracked older-appearing brain profiles, with SAGs aligning with structural, functional, and combined brain age gaps in healthy controls, AD, nldFTD, and ldFTD. MCI could not be included because the necessary paired data were unavailable.

SAGs also aligned with epigenetic aging. Associations were found across all three DNA methylation clocks in healthy controls and AD, and across two of the three in ldFTD.

Speech clock captures more than individual features

When comparing healthy controls with the dementia groups, SAGs provided stronger discrimination than individual speech-feature domains or a combination of the strongest linguistic and acoustic features. Linguistic and semantic-memory features were particularly informative, especially in ldFTD.

Brain-network analyses added to the picture: SAGs were more strongly associated with whole-brain aging than with individual functional networks, although smaller links emerged with language-related and salience networks. The authors suggest that speech may therefore act as a “functional expression of systemic aging,” capturing information that converges with cognitive, biological, and social dimensions of aging.

Limitations

Despite the study's strengths, it has several limitations. It is cross-sectional and so precludes causal inference, the identification of aging trajectories within each individual case, or prediction of future changes in category. In particular, a positive SAG represents an older-appearing speech profile at the time of assessment rather than evidence that an individual is aging more rapidly over time. Longitudinal studies will be necessary to validate these findings and establish their predictive utility, and to follow changes over time.

Residual noise may have persisted in the linguistic features because recordings were obtained at different locations and centers and under different conditions. Automated tools might have introduced biases and require further assessment. Sex, medical comorbidities, and mood-related fluctuations may also influence speech and contribute to variability in SAGs.

MCI and ldFTD subgroups were of limited size, reducing statistical power, especially to detect small effects. Biomarker and exposome data were missing for some participants and may have biased some analyses.

Participants were all Latin American, used structured speech tasks, and the findings may not generalize to other languages. Future studies should use other languages and cultures with more natural-appearing language tasks to validate these findings and their generalizability.

Speech emerges as a potential window into aging

SAGs provide a biologically meaningful, scalable, and culturally adaptable measure of aging that aligns with cognitive performance, fluid biomarkers, social exposome, neuroimaging-derived brain clocks, and epigenetic aging.

If confirmed, speech clocks could eventually offer a scalable proxy for aspects of aging in research and potentially clinical settings, and help inform equitable public health measures to improve dementia prevention and promote brain health.