A machine learning-based “speech clock” analyzes hundreds of speech parameters to measure brain, cognitive and social aging.
There is a constant in our days that tells the story of how we age much more faithfully than wrinkles and gray hair: the voice. The language we use, the choice of words, intonation, speech pauses and hundreds of other parameters, analyzed by machine learning systems, can form a “voice clock” that faithfully reveals whether the age of our organism is in line with that written on the documents. This is what emerges in a study published in Science Advances.
Sieve conversations
A group of scientists from Trinity College Dublin analyzed with a machine learning system the spoken language of almost 3000 Spanish-speaking adults from Argentina, Chile, Colombia, Mexico and Peru, both healthy and suffering from mild cognitive decline, Alzheimer’s or forms of frontotemporal dementia.
The algorithms used are able to “learn” recurring patterns in the data fed to them, and to formulate accurate inferences about new data based on what they have learned. They analyzed hundreds of acoustic and linguistic characteristics of speech, such as speed and pauses, voice inflection, the emotional content of language, the choice of words within sentences, the precision in conveying meaning and the organization of verbal production.
The models made it possible to estimate the chronological age of the participants starting from these characteristics, but also to formulate the vocal gap for each individual, i.e. the difference between the real age and the age estimated from listening to the voice.
Weight differences
Problems, in fact, arise if there is a significant gap between the real age and that estimated from speech: those who had a very “mature” language compared to their chronological age showed more evident signs of various types of aging.
- He was subject to accelerated brain aging, as identified by neuroimaging tests.
- He had a faster biological clock, as confirmed by three different types of DNA methylation tests, which measure chemical changes in gene activity related to advancing age.
- He showed a more advanced cognitive age: larger gaps between speech clock and actual age were associated with greater difficulties in executive functions (such as planning, organization, regulation of behaviors) and memory.
Traces of dementia in his voice
The progressively larger gaps between voice clock and chronological age were also associated with pathological aging, i.e. Alzheimer’s and other forms of dementia. In Alzheimer’s patients, the larger gap was correlated with higher levels of plasma p-tau217, an important biomarker of the disease.
This does not mean, for the avoidance of doubt, that voice tests can “predict” the onset of dementia or measure its progression: the study is in fact of a transversal type, it is like a group photo which measures a series of parameters in a given population at a specific moment in time, and not over a series of years.
The voice of lonely people
The accelerated aging of voice and language is also linked to each person’s social ecosystem: the presence of healthy relationships, education, a favorable economic situation, access to healthcare, food security, a peaceful childhood behind us.
Listen to heal
The potential of this tool lies in its ease of access: these voice analysis systems are cheap and low-cost, implementable in low-income countries and underrepresented in scientific studies. This type of tool could help measure the aging trajectory of the population in contexts where expensive molecular analyses, blood biomarker searches or magnetic resonance imaging are not always or not available to everyone.
