Giving Computer Voices a 'Human Touch'
Companies Deliver Personalization with Friendly, Helpful Machines
Listen to David Kestenbaum's report.
April 18, 2002 --
Call a car rental agency, an airline or a movie theater, and you're likely to spend the entire call talking to a deceptively human computer voice. The voices behind the latest generation of automated phone-answering systems refer to themselves in the first person, and they engage callers in something that can feel almost like a real conversation. They're an example of a new relationship between human and machine -- a relationship some researchers think is a mistake.
People who have talked with these systems say they find themselves liking the voice on the other end -- possibly even infatuated with it. The systems are designed to be engaging, so it's not so crazy to want to talk to the machines, says Blade Kotelly. He works for a company called SpeechWorks, which makes speech recognition systems. One reviewer wrote, "after this kind of system you will want to take the voice to dinner."
Only a few years ago, customers had to push buttons to interact with computerized phone systems, and speech recognition only worked if you spoke the Queen's English into a high quality microphone. But voice recognition software has become more clever, as have designers. The conversations are carefully choreographed so the computer only has to understand a handful of statements.
During the design process, Kotelly and others map out hundreds of possible conversational paths on a flow chart. They think about the pauses, the cadence, and the remarks that make a conversation sound fluid, like an "oh by the way." Kotelly argues that all this effort to give a computer personality is worth it. The interaction goes more smoothly, and people feel better about it.
Yet some people get a little queasy when a computer says "I" and puts on human airs. Ben Shneiderman, a computer scientist at the University of Maryland's Human-Computer Interaction Lab, says most people like machines to act like machines.
"I think they want to know that they are different from a machine," says Shneiderman. "The deception that a machine could be like a person is troubling to most people."
For All Things Considered, NPR's David Kestenbaum has more.
The University of Maryland Human-Computer Interaction Lab
More demos from SpeechWorks
More demos from General Magic