StressSense -- SONVB research in ACM TechNews

Monday, August 27, 2012 The StressSense system was featured in ACM TechNews.

Computer scientists have trained a system to detect stress in a person's voice. StressSense is designed to initially recognize someone's unstressed voice, then compare this recording to preprogrammed knowledge of stress-caused physiological changes, such as faster speech and clipped frequency spectrum. The prototype has a stress-recognition accuracy of 81 percent indoors and 76 percent outdoors. "We propose StressSense for unobtrusively recognizing stress from [a] human voice using smartphones," the researchers say. They note the StressSense classifier "can robustly identify stress across multiple individuals in diverse acoustic environments." The software has the potential to raise user awareness of stressful events and help them cope, says Intel's Hong Lu. The researchers demonstrated that stress from a human voice can be recognized using smartphones in indoor and outdoor conversational data, and that a universal stress model can be adapted to specific individual users. Furthermore, a stress model can be adapted to unseen environments, reducing the cost of training stress models for different situations, and the proposed stress classification pipeline can operate in real time on Android smartphones.  

Link to original article:
Voice-stress software is put to the test
by Nancy Owano
August 22, 2012

SONVB project