Iulia Lefter of Delft University of Technology and colleagues at the Netherlands Defence Academy and TNO Defence, Security and Safety, explain how emergency call centres are commonly overwhelmed by the sheer number of calls, especially during disaster situations or other national emergencies. A system that could distinguish automatically between a seriously urgent call and a more mundane issue could reduce the burden considerably and allow calls to be prioritised more effectively.

"Stress and negative emotions, in general, have a strong influence on voice characteristics," the researchers explain. "Because speech is a natural means of communication, we can utilise the sound patterns of speech to detect stress and (negative) emotions in a non-intrusive way by monitoring the communication." Factors such as how quickly a person is talking, whether or not there are rises and falls in pitch and tone and breathing rate, all change when we are stressed and can be detected.

The team has now "trained" a computer algorithm that receives audio input from emergency calls to assess the emotive level of the callers' speech. Four different training techniques were used with recordings from actual emergencies of known outcome and the team says their error rates are as low as 4.2% for a database of call centre recordings used in the research. Optimisation of the algorithm using a larger training set and more robust statistical tools might improve that still further.

The researchers expect the system to have military applications in the first instance. However, it could be adapted to the civilian emergency services and perhaps other applications, such as criminal investigations.

Note: If no author is given, the source is cited instead.

• Perception
• Language Acquisition

• Detectors
• Acoustics

• Disaster Plan
• Resource Shortage

• Videoconferencing
• Voice over IP
• Sound effect
• Emotional detachment
• Psychological trauma
• Communication