Researchers Develop AI Model That Detects Cognitive Impairment from Speech
Researchers at Immanuel Kant Baltic Federal University have developed a neural network that analyzes speech recordings to identify signs of cognitive impairment, including declines in memory, attention, and cognitive performance.
The model evaluates 88 acoustic features, including pauses, voice timbre, loudness, and vocal tremor. It was trained on recordings from people with clinically diagnosed cognitive impairments as well as healthy volunteers. Rather than analyzing only what a person says, the system also evaluates how they speak. It does not make a medical diagnosis but instead estimates the probability of cognitive impairment on a scale from 0% to 100%, making it suitable as a clinical decision-support tool for rapid screening.
According to Osadchy, the model compares a speaker's characteristics with patterns derived from both healthy individuals and patients with diagnosed cognitive disorders, identifying statistically significant indicators of impairment. The research was presented at a conference in Novosibirsk.
