The Art and Business of Speech Recognition: Creating the Noble Voice

Flight information systems were one of the earliest types of commercial speech-recognition applications. The United Airlines system has been in operation since 1999 and receives an average of two million calls per month. Between 1999 and 2002 this system saved United Airlines in excess of $25 million over the touchtone system that previously was in place. ^[1] While business travelers are an airline's most profitable customers, they are not their sole customers, so it only makes sense that the United Airlines system provides expedited flight information for those callers who know their flight numbers as well as those who don't. Here's how a typical call goes.

^[1] Bob Bongiorno, Managing Director of Customer Service, Planning and Finance Applications, United Airlines, at SpeechWorks International Global Speech Day "Web Seminar," May 22, 2002.

This system was designed with two calling populations in mind. The primary audience is frequent-flier business travelers. Because these callers use the system frequently ”and are always in a hurry ”they're interested in learning how to get quick responses to their inquiries. This system teaches them that they'll get the fastest service if they provide the flight number ”either by saying it or by entering it on the telephone keypad. After that, all they need to do is specify whether they want arrival or departure information, and the entire call can go very quickly. In fact, callers can make the call even shorter by "barging in" ”interrupting the initial prompt by saying or entering the flight number.

The other people who call flight information systems are friends , relatives, and colleagues checking to see when they need to pick up their parties at the airport. Often these people are unaware of the flight numbers, and may not even know the exact time of the scheduled arrival. So, a very different type of experience is needed for these callers ”hence, the use of the "I don't know it" command.

The "I don't know it" command mimics the interaction a person might have with another person. We could use another prompt, such as "Enter or say the flight number or say 'Find the flight number' if you don't have it," but that would be cumbersome and unnatural in this context ”particularly considering that millions of people of all ages call into the system.

Also note the use of the word "it" in the command ”"I don't know it. " There's a reason why "it" is there. Although some people might use an abbreviated command, such as "I don't know," that doesn't make sense grammatically, because it doesn't specify exactly what the caller doesn't know. Try saying these two examples out loud and compare the way they sound.

Or:

The second one just doesn't sound right. And would anyone think that they're reducing the time of the call by omitting the word "it?" In fact, recognizers are more accurate when the utterances are longer than shorter. And while the "it" sound in this example won't improve recognition accuracy because it doesn't change the overall sound of the utterance considerably or add much to the length, in certain contexts small additions can make a significant difference.