Microsoft researchers are testing a new feature for Teams that aims to provide speakers on calls with a near-real-time assessment of the moods and reactions of their audience.
The ‘AffectiveSpotflight‘ is said to be built from a type of facial recognition algorithm that uses a neural network to capture and assess the expressions of call participants, monitoring for changes in emotions such as happiness, sadness and surprise.
The software is being developed by researchers across a number of Microsoft facilities in Redmond, Boston and Cambridge, MA, with findings expected to be revealed at Japan’s CHI Conference on Human Factors in Computing Systems in May.
The system is said to be able to spot subtle movements, such as the shake of a head, a furrowed brow, and even a raised eyebrow. Each of these is then rated between 0 and 1, with positive emotions scoring higher. The person with the highest score is highlighted to the presenter, every 15 seconds.
The facial expressions of the participants are also matched to datasets in Microsoft’s Convolutional Neural Network (CNN), which has expression categories for anger, disgust, fear, happiness, sadness, surprise, and neutral.
«Public speaking is often regarded as one of the most stressful daily activities and is heavily influenced by audience responses to the presenter,» the research states. «In fact, studies that seek to reliably induce acute stress on people often involve giving a presentation in front of a neutral-looking audience (a.k.a., Trier social stress test). While research on audience responses in online settings is still nascent, there is prior work considering the impact of in-person audience responses, especially in the context of alleviating public speaking anxiety.»
The feature isn’t available on Microsoft Teams as yet, but it is very much in keeping with recent updates to the platform that focuses on wellbeing and combating so-called ‘video call fatigue‘.
However, this may be seen as a somewhat overly technical solution to a problem that is fairly easily solved with feedback, and it isn’t difficult to imagine how this feature could create further anxiety as it tries to reduce it.