AI System in Russia Detects Anxiety Through Voice Analysis
Artificial intelligence can detect subtle vocal changes that the human ear may miss.
Researchers from the Department of Cyberpsychology at the Faculty of Social Sciences of Lobachevsky State University of Nizhny Novgorod are developing machine learning models capable of identifying anxiety from a person’s voice, the university’s press service reported.
According to Valeria Demareva, PhD in psychology and head of the Department of Cyberpsychology at the university’s Faculty of Social Sciences, automatic detection of stress and anxiety through voice analysis can help identify vulnerable conditions among operators, dispatchers and medical professionals at an early stage.
When a person becomes nervous, muscle tension increases and breathing becomes faster. As a result, the voice may sound harsher or begin to tremble. Timbre, volume and speech tempo also change. Artificial intelligence can detect these micro-variations, even when they are too subtle for the human ear to notice.
Recognition Accuracy Reaches 92%
Researchers compared recordings of students speaking under two conditions: one group delivered a presentation in public before a panel and fellow students in a lecture hall, while the other spoke privately in a quiet office without an audience. The team extracted acoustic features and evaluated the differences between the two sets of recordings.
After careful signal cleaning and extraction of Mel-frequency cepstral coefficients (MFCC), a Gradient Boosting machine learning classifier was able to distinguish anxiety in speech with an accuracy of 91.9%.
