The app called 'StressSense' is first trained to recognise a person's unstressed voice.
For that, users must relax and read a 3-minute passage from a book into their phones.
The system then compares this recording to its pre-programmed knowledge of the physiological changes that stress induces like speaking at a faster rate and a clipped frequency spectrum.
The application then takes note of any instances of stress it detects in the voice.
"Our stress model also adapts to different background noise environments," New Scientist quoted Hong Lu of Intel in Santa Clara, California, who developed the system, as saying.
In tests that included putting volunteers through mock job interviews, the researchers found their prototype's stress-recognition accuracy to be 81 per cent indoors and 76 per cent outdoors, where sound quality wasn't as good.
The team plans to make the system a plug-in to an Android application called BeWell, which uses a phone's accelerometers and Global Positioning System (GPS) sensors to record users' activity and sleep levels.
More From This Section
Smartphone users will be able to set StressSense to either listen to their voice throughout the day, or only to activate when they are having a phone conversation.
The app will be presented at the Ubicomp conference in Pittsburgh, Pennsylvania, next month.