Automated voice response systems can be maddening, especially when you can’t get them to understand your speech or the question you want answered. A team at the Universidad Carlos III de Madrid and the Universidad de Granada is trying to make them less annoying -– and contribute to a reduction in the number of curses screamed into telephones worldwide.
What they did was build a system that recognizes the emotional state of the user by looking at conversational cues. But to do that they had to figure out what the cues were.
NEWS: 7 Formerly Popular Sites That Are Dying
The study, by Professors David Grill Barres, of the Applied Artificial Intelligence Group of UC3M’s Computer Science Department, Zoraida Callejas Carrion and Ramon Lopez-Cozar Delgado, of the Computer Languages and Systems Department of the UGR, focused on negative emotions that can make talking with an automatic system frustrating. The three emotions they focused on were anger, boredom and doubt.
To automatically detect these feelings, they gathered data on 60 different aspects of the sounds people make, including tone of voice, speed of speech, the duration of pauses and the energy of the voice signal.
They also looked at how the dialogue developed, that is, what direction the conversation was taking. That information helped adjust the odds that a user was in one emotional state or another. For example, if the system did not correctly recognize what the person taking wanted to say several times, or if it asked the user to repeat information, these factors could anger or bore the user when interacting with the system.
Predicting the dialogue is important; one of the annoying things about dealing with machines is that they seem oblivious (because they are) that repeating oneself is irritating. So the team developed a statistical model using earlier dialogues to build a picture of what users are likely to say and what state they are in when they do.
NEWS: Meet 'Robonaut 2,' Your Future Co-Worker
Once the machine has picked up not only on the users emotion (knowing they are, say, angry) but what their likely intention is, the scientists propose automatically adapting the dialogue to the situation. A person who has doubts needs more help, while a bored person (or one that is annoyed) doesn’t want that. The team built the guidelines by testing the system with real humans, and was able to show that it made the “help desk” calls shorter.
Of course, a real benefit might be lowering the worlds collective blood pressure; perhaps another area of study will be on long term health of populations where this is adopted.
Image: Wikimedia Commons/ Ben Schumin
comments ( )