A team of researchers from Delft University of Technology, the Netherlands Defence Academy and TNO Defence, Security, and Safety are working on a little black box which can distinguish just how much of an emergency a caller has. In other words, you call 911 to report that your house is on fire, and a 'trained' computer figures out how desperate you are.
Basic model for speech emotion recognition: MSc Thesis, Iulia Chiriascescu, Delft University of Technology
The emotion detector, which does not have an official name yet, is the extension of lead author Julia Lefter's dissertation for Delft two years ago. Now, taking the emotion detector very seriously, the Netherlands defence authority and a defence contractor are developing it for use during disaster situations or national emergencies.
The team has a trained computer algorithm that receives audio input and assesses the emotive output of the call. The emotion detector was tested on a data base of actual calls made to emergency centers and its error rates were only 4.2 percent.
Having some experience with voice analysis technology, I can appreciate the emotion detector's capabilities. Voices are extremely sensitive to stress and negative emotions, and voice analyzers can pick up the tones and pitches of speaker. Additionally, breathing rate and the rate of speech can be measured to determine stress very accurately.
A larger database of calls will be used to refine the emotion detector's abilities even more. Though the first applications will be for military use, it is possible that it will be used for civilian emergencies and even for criminal investigations in the future.