At a surface level, the sound of someone’s voice can offer insight into their mood and stress levels. But many medtech developers are aiming to go even deeper, applying artificial intelligence to snippets of speech to pick up on potential signs of disease.
So far, these vocal biomarkers have been linked to cardiovascular and respiratory conditions including COVID-19, as well as neurological diseases like depression, anxiety and even dementia—the last of which is in the crosshairs for speech analysis AI developer Winterlight Labs.
Winterlight has built a digital biomarker platform that automatically combs through speech samples to detect and track the progression of dementia. Results of a new analysis performed in collaboration with Genentech prove the technology’s ability to monitor Alzheimer’s disease with similar accuracy to standard neuropsychological tests, the AI maker announced this month.
The analysis relied on recordings gathered from 101 early-stage Alzheimer’s patients already enrolled in Genentech’s Tauriel trial, which was studying whether the anti-tau antibody semorinemab can slow the progress of the neurodegenerative disease.
The recordings were captured during standard administrations of the Clinical Dementia Rating test, which relies on interviews with patients and their caregivers to assess six areas of cognitive and behavioral health: memory, orientation, judgment and problem solving, community affairs, home and hobbies performance and personal care.
Winterlight’s digital biomarker platform analyzed the recordings and was able to generate more than 500 distinct acoustic and linguistic markers. Nine of those were pinpointed as being strong indicators of neurological decline—including some related to word length, word frequency, the voice’s power spectrum and more—and were subsequently combined into a new scoring system for Alzheimer’s progression.
When compared to the clinician-administered assessments that are typically used to track the progress of dementia, Winterlight’s automated analysis churned out either similar or even more accurate ratings. For example, according to the company, its AI gave a better indication of longitudinal change than the Repeatable Battery for the Assessment of Neuropsychological Status test and the language-focused parts of the Alzheimer's Disease Assessment Scale-Cognitive Subscale test.
Those promising results mean that AI could be used not only to strengthen doctors’ assessments of dementia patients, per Jessica Robin, Winterlight’s director of clinical research, but also to improve clinical trials of Alzheimer’s treatments.
"Given the crucial role that language changes play in Alzheimer’s, automated language processing represents a new tool to characterize speech and language patterns and provide additional insight into a patient’s condition,” Robin said in a press release. “We’ve worked to streamline the process of implementing automated tools into research, allowing for low-burden assessments suitable for remote testing that can help demonstrate how language changes because of disease progression or therapeutic interventions.”
Still, the researchers noted, further validation is needed to confirm the relevance of the speech-based, fully automated assessment in tracking the disease.