A person’s speech may reveal more than what they are trying to say. Changes in pauses, speaking speed, pitch, vocabulary and the precision of language can reflect processes linked to cognitive decline and aging, according to a new study published in Science Advances.
Researchers developed a machine-learning “speech clock” using recordings from 2,928 Spanish-speaking participants in Argentina, Chile, Colombia, Mexico and Peru. The group included cognitively healthy people and participants with mild cognitive impairment, Alzheimer’s disease, and language-dominant and non-language-dominant frontotemporal dementia.
The model estimated a person’s age from acoustic and linguistic features. Researchers then compared that estimate with the participant’s actual age, creating a “speech age gap.” A positive gap indicated a relatively older-sounding speech profile, while a negative gap indicated a relatively younger or more preserved profile.
Speech age gaps were lowest among healthy participants and higher across the patient groups. The largest gaps were observed in people with language-dominant frontotemporal dementia, followed by non-language-dominant frontotemporal dementia and Alzheimer’s disease. The differences remained significant after the researchers adjusted for factors including age, sex, education, country, comorbidities and number of languages spoken.
Speech and biological aging
The speech measure was also associated with several independent indicators of aging and neurodegeneration. Larger speech age gaps were linked to poorer cognitive and functional performance, higher plasma levels of p-Tau217—a biomarker associated with Alzheimer’s disease—and more adverse social-exposome profiles.
The researchers also compared speech age gaps with brain-age estimates derived from structural and functional MRI and with DNA-methylation aging clocks. Participants whose speech appeared older for their chronological age generally showed larger brain-age gaps and more advanced epigenetic aging. These associations suggest that speech may reflect overlapping biological, cognitive, motor and social processes rather than acting only as a marker of language ability.
The strongest speech predictors included measures of verbosity and speech timing, as well as pitch and emotional content. Longer pauses, slower speech and changes in word choice can reflect difficulties with processing speed, word retrieval and semantic memory, although these features can also be influenced by education, mood, culture, hearing, illness and the recording environment.
A possible screening tool, not a diagnosis
The attraction of a speech clock is practical. Speech recordings can be collected remotely and analyzed automatically without MRI, blood testing or specialized equipment. This could make speech-based measures useful for research and potentially for large-scale monitoring, particularly in communities where advanced diagnostic resources are limited.
However, the study does not show that a speech recording can diagnose dementia or predict with certainty who will develop it. The research was primarily cross-sectional, meaning that participants were assessed at one point rather than followed over time. It therefore cannot establish whether a larger speech age gap represents a progressive aging process or predict future cognitive decline.
The model was also developed using Spanish-speaking participants from five Latin American countries. Its performance in other languages, cultures, dialects and less structured conversations remains to be established. Differences in microphones, recording conditions, education, multilingualism and speech prompts may also affect the results.
The findings point to speech age gaps as a promising research biomarker—one that could complement clinical assessments, blood tests and brain imaging rather than replace them. Larger longitudinal and multilingual studies will be needed before speech clocks can be evaluated for routine clinical use.
Source: Science Advances
