Scientists are developing an "emotion sensor" to show if someone is finding your conversation interesting or not.
The computer would detect this man was not interested
It is being developed to help people with autism, who tend to be less skilled at interacting with others.
New Scientist magazine reports researchers at the Massachusetts Institute of Technology (MIT) have developed the headset.
A camera on a pair of glasses is linked to a hand-held computer which "reads" the emotional reactions of a listener.
The device uses image recognition software and emotion-reading software to decode the images.
If the wearer appears not to be engaging with their listener, the software makes the computer in their hand vibrate.
Previous research by the team has shown the device could detect if someone was agreeing, disagreeing, concentrating, thinking, unsure or interested from just a few seconds of film.
Previous computer programmes have only been able to detect six basic states of happiness, sadness, anger, surprise and disgust.
The MIT system was "primed" with 100 clips of actors displaying particular emotions.
It detects movements of the eyebrows, lips and nose and tracks head movements such as nodding, shaking or tilting.
In recent tests, the device correctly identified people's emotions in video clips 90% of the time.
When it was tested with members of the public, it was right 64% of the time.
The team are about to begin the first tests of day-to-day use with volunteers, some of whom have autism.
The system in its current form takes a lot of computer power so may need to be pared down to be manageable.
For common use, the headset will have to have a high-resolution digital camera which can be worn comfortably.
And people with autism would need to be trained to look at the faces of people they are talking to so the camera can pick up expressions.
Professor Simon Baron Cohen of the Autism Research Centre in Cambridge, said: "It is a very clever application as a prosthetic device.
"The success of the project will depend on how accurate the software really is, whether the feedback to the human user is fast enough to be useful in real-time social situations, and how unobtrusive it is."
Timothy Bickmore, of Northeastern University, who studies ways in which computers can be made to engage more with people's emotions, says the device could be particularly useful in the classroom.
"I would love it if you could have a computer looking at each student in the room to tell me when 20% of them were bored or confused."