Computer scientists at the University of Rochester have created an artificial intelligence (AI) tool that can help detect early signs of Parkinson’s disease just by analyzing how someone speaks.
The tool uses voice recordings from people reading two short sentences—called pangrams—that contain all 26 letters of the alphabet.
The AI can then analyze these recordings for signs of the disease in seconds with nearly 86 percent accuracy.
Parkinson’s is the world’s fastest-growing neurological disorder. It is usually diagnosed by movement disorder specialists using medical history, physical exams, and brain scans.
However, this new AI tool is not meant to replace a clinical diagnosis. Instead, it offers a quick and easy way to screen for Parkinson’s and could be especially useful for people in areas without access to neurological specialists.
Professor Ehsan Hoque, who co-led the study, said, “Many areas around the world lack specialized care for Parkinson’s. A tool like this could help people know when it’s time to seek further testing.”
Hoque and his team suggest that with users’ permission, voice-based technologies like Amazon Alexa or Google Home could help detect potential cases of Parkinson’s.
To build and test the tool, the researchers collected voice data from over 1,300 people, both with and without Parkinson’s. These recordings were made in different places, such as people’s homes, hospitals, and care centers.
Participants were asked to read aloud the following two sentences: “The quick brown fox jumps over the lazy dog. The dog wakes up and follows the fox into the forest, but again the quick brown fox jumps over the lazy dog.”
Using advanced speech models trained on millions of audio clips, the AI learned how to recognize small changes in how words are spoken—such as differences in tone, pauses, breathing, and unclear sounds—that are often found in people with Parkinson’s.
Lead author Abdelrahman Abdelkader explained that the way people with Parkinson’s say these sentences often includes small signs, like trailing off or unusual pauses. These patterns are different from how people without Parkinson’s speak, and the AI can detect these subtle differences.
The AI achieved 85.7 percent accuracy in identifying whether someone might have Parkinson’s. While the tool focuses on speech, Parkinson’s can also show up in other ways, such as changes in movement or facial expressions.
For years, Professor Hoque’s lab has worked on combining various signs to improve screening tools. Their goal is to develop easy-to-use technology that can detect Parkinson’s early and encourage people to get medical help.
Tariq Adnan, another lead author and PhD student in Hoque’s lab, noted that about 89 percent of Parkinson’s patients have some form of voice abnormality, making speech a strong first step for screening. The lab also offers interactive demos online that combine speech with other tests to increase accuracy.
Other contributors to the study include PhD students Md. Saiful Islam, Zipei Liu, Ekram Hossain, and Sooyong Park. The project was funded by the National Institute of Neurological Disorders and Stroke, the Gordon and Betty Moore Foundation, a Google Faculty Research Award, and a Google PhD Fellowship.
If you care about Parkinson’s disease, please read studies about how to improve walking in people with Parkinson’s disease and Scientists find causes of Parkinson’s and Lewy body dementia.
For more about Parkinson’s disease, please read studies These common drugs may increase risk of Parkinson’s disease and Researchers find an important cause of Parkinson’s disease.
The study is published in npj Parkinson s Disease.
Copyright © 2025 Knowridge Science Report. All rights reserved.
