13/05/2026
Computers that can detect and even respond to human emotions may seem like the stuff of science fiction. But the potential, especially in mental health care, is enormous.
Imagine a wearable device that monitors your emotional state, using brain sensor technology to detect early signs of stress, anxiety or low mood before you’ve even noticed them yourself. That kind of technology could help people manage conditions like depression or burnout, or alert healthcare providers when someone needs support.
New Auckland Bioengineering Institute graduate Alireza Farrokhi Nia’s PhD has taken a step closer to that goal. He has developed a system that avoids facial expression or tone of voice (signals that people can easily hide), instead going straight to the source: the brain.
Unlike other research in this space, Reza’s work combined two types of brain-monitoring technology: EEG, which captures electrical activity, and fNIRS, which measures changes in blood oxygen levels.
He recorded volunteers’ brain signals while they listened to music, everyday sounds (parents fighting, glass shattering, a running river), and watched videos (including funny cat compilations, the joyful moment a deaf baby hears her sister’s voice for the first time, news footage of famine, and neutral clips like a barbershop haircut) designed to trigger a full range of emotional responses.
Reza says he wasn't trying to measure whether someone felt ‘happy’ or ‘sad’ in any simple sense – human emotion is too complex for that.
Instead, he measured emotion along broad dimensions based on how pleasant the feeling was and how intense.
Collecting the data was a huge undertaking. More than 100 people took part, each spending around two hours in the lab. Every detail had to be controlled from lighting to the words spoken to participants, even down to Reza wearing the same cologne for months so nothing unintentionally influenced the subjects’ mood.
Once the recordings were complete, the real challenge began: building AI models that could interpret the brain signals. Harder still, he needed models that worked across different people, not just one individual.
Read more at abi.ac.nz