There’s a lot of brain power that goes into speech. First, there’s a thought or idea, which the brain must translate into words. Those words are then translated into specific movements of the lungs, tongue and mouth to shape the sounds. Those movements then must be perfectly executed and timed with breath. If there is damage to the brain from a stroke or a brain disease, the timing of the movements or translation can go wrong.
Because of that, changes in voice and speech can provide the first clues to a neurodegenerative disease. Voice samples collected for research can help diagnose neurogenerative diseases early, explains Hugo Botha, M.B., Ch.B., a behavioral neurologist and associate director of the Neurology Artificial Intelligence Program, at Mayo Clinic in Rochester, Minnesota.
“There are some diseases where the very first manifestation is in someone’s voice or their speech,” Dr. Botha says. Those include Parkinson’s disease; atypical parkinsonism such as multiple system atrophy, progressive supranuclear palsy and corticobasal syndrome; amyotrophic lateral sclerosis (ALS); myasthenia gravis; and some types of frontotemporal dementia that can result in aphasia.
As part of clinical practice, Mayo Clinic’s neurology patients often are recorded when they have their voice or speech examined, allowing clinicians to track the disease over time.
“But separate from the clinical practice, we have a large research program at Mayo, where we are collecting voice and speech samples using an application that runs on the person’s phone or the laptop computer,” Dr. Botha explains.
To collect the voice samples, patients are tasked with running through a series of exams remotely.
“They could do it — say every couple of weeks, every couple of months — so we can really get a longitudinal view of their disease instead of just a snapshot,” says Dr. Botha says.
The creation of this large and growing speech bank, which securely stores all speech and voice samples, can be used for research, including using it to train artificial intelligence (AI) algorithms.
“There are some signals in someone’s voice and speech that a computer or an algorithm might pick up on, that a human listener wouldn’t pick up on. And so that’s more of the sort of research, AI side of things, where we’re trying to use hundreds of recordings and patients with various diseases, and then trying to see if the computer can separate those diseases, even though human listeners may not be able to,” Dr. Botha says.
