Posted: 12/1999
The Name Game
Speech Recognition Ushers DA into 2000
By Liz Montalbano
Directory assistance (DA) has come a long way since early telephone users relied on operators to connect them to the party they were calling. In recent years, time spent talking to a live operator when a user dials "411" in the United States has become significantly reduced, thanks to systems that automate the front end of DA calls. This automation trend is only expected to expand further with the development of speech recognition-enhanced DA.
While nationwide systems for speech recognition-enhanced DA are not deployed yet, there are applications in the works that possibly will make about 75 percent of DA calls entirely automated through speech recognition, says Kathy Frostad, director of telecommunications product marketing, Nuance Communications, Menlo Park, Calif.
"In the U.S. today there's about 8.5 billion [DA] calls made annually, both wireless and wireline," Frostad says. "About 80 percent of those calls are business and government. When you look at what can be automated, we believe that with full system deployment over six to nine months, we could achieve automation of easily 75 percent or more of the calls."
In fact, according to research by Computer Economics Inc., Carlsbad, Calif., the electronic voice market itself will experience a $4 billion growth by 2005, with an increase in the voice recognition market from 300 million this year to 1.2 billion in 2005 (see chart, "Electronic Voice Market Growth, 1999-2005," below).
Source: Computer Economics Inc., Carlsbad, Calif.
Nuance currently develops speech recognition-enhanced applications that are installed in equipment made by partners such as Telcordia Technologies Inc., Morristown, N.J., and Motorola Inc., Schaumburg, Ill. Businesses such as Fidelity Investments, Boston, and Charles Schwab & Co. Inc., San Francisco, use Nuance applications to provide customers with stock quotes and to support mutual funds trading. Nuance also develops speech recognition-enhanced auto attendants for use in company employee directories. Building upon such directory deployments, in August Nuance demonstrated an application for speech recognition-enhanced nationwide DA supporting listings for 100 million businesses and residences.
"We wanted to prove our capability to handle the complex application of directory assistance and we've done that with the demonstration," Frostad says.
She adds that Nuance currently is meeting with third-party DA providers and carriers that want to provide their own DA to decide how to implement wide-scale deployment of speech recognition-enhanced directory services.
"I'd say each deployment's probably going to take about nine months, so it won't be until about the mid-half of next year that we'll see live deployments up and running," Frostad says.
The Road to Speech Recognition-Enhanced DA
The road leading from live operator-assisted DA into the future of automated systems is anything but straight and narrow. Frostad maintains that "the hardest one of these telecommunications applications [to make speech recognition-enabled] is directory assistance," mainly because of the sheer volume of listings a system must be able to recognize. While companies at the forefront of speech recognition technology may be gearing up to provide it for nationwide DA, it hasn't been an easy task.
Mark Bannon, vice president of sales, marketing and technical services for Phonetic Systems, Burlington, Mass., says his company's phonetically based speech recognition engine "took seven years to develop," mainly because of the infinite possibilities for name pronunciation.
"Look at a person's business card and figure out all the different ways in which you could actually say that name," he says, "and if you're going to be an auto attendant product like ours, you're going to have to respond to different pronunciations."
Bannon points out there is a difference between current automated DA systems and actual speech recognition-enhanced systems, which often are mistaken for similar creatures.
"Did you ever call the phone company and get what you think is an automated system answering the phone and asking for your listing?" he asks. "Do you know that that's a person?"
Bannon goes on to explain that current DA systems use store-and-forward recorded messages backed up by actual agents who listen to the request and send the listing name on to a DA operator once one is available. This is a far cry from the technology that companies such as Phonetic Systems and Nuance are working on, which actually will recognize what the caller is saying and will eliminate the need for a front-end agent.
While current automated attendant systems such as those manufactured by Nuance; Phonetic Systems; Philips Voice Request, a division of Philips Electronics, Amsterdam, the Netherlands; and InterVoice-Brite Inc., Dallas, are the most immediate precursor to a widespread deployment of speech recognition-enhanced DA, Frostad says Nortel Networks, Richardson, Texas, was one of the initial innovators of the technology, providing platforms that would automate the front end of calls.
"They were just doing the front end of the call, automating, 'Directory assistance, how may I help you?' [and] 'What city and state?'," she says. "So you'd automate recognizing the city and state and if [the system] couldn't recognize it then [it] would default to an operator."
She says that eventually, early speech recognition applications for DA could recognize frequently requested business listings, but only in about the top 4,000 listings, "which really isn't that many when you consider all the millions of listings that are out there," she adds.
But according to Bannon, the speech recognition technology that companies like his are developing is extremely sophisticated, focusing on the phonetics of language rather than an actual name match.
"We take a directory and we look at the directory and we create a phonetic cloud of potential pronunciations around a textual representation of a name," he says. "Then we try to match what the caller says against those phonetic clouds, so if somebody's mispronouncing or if there's alternate pronunciation for names, we're going to be able to find that person in there."
The investment in such technology has paid off. In September, Phonetic Systems launched a global DA technology called Directory-Assistant for databases with more than 500 million listings. And while Bannon says his company--like Nuance--does not actually have any commercial deployments right now, "you'll see these systems by ourselves and others by the end of next year."
Benefits of Speech Recognition-Enhanced DA
Companies working fervently to develop applications to enable large-scale speech recognition-enhanced DA aren't doing it for their health--they're trying to save DA providers and carriers money. Lots of it.
Bannon says that because of the high volume of calls made to DA every year, the infrastructure and labor costs of providing live operators takes quite a chunk out of a carrier's or DA provider's revenues.
"There's so many calls made on DA ... there's so much money spent on infrastructure there," Bannon says, "that every second the phone company saves translates into 70 [billion] or 80 million [dollars] in savings."
Frostad concurs, estimating that the cost of fully automated systems is about one-tenth the cost of operator-supported DA. Since operator-assisted calls range in cost from 25 cents to 50 cents per call, and local telephone companies have a maximum limitation to charge about 30 cents to 35 cents per call, there is little revenue in providing DA service.
On the other hand, "You're looking at a four-to-12-month payback [using speech recognition-enhanced DA]," Frostad says.
Frostad cites two other reasons carriers and DA providers might want to consider implementing speech recognition in their DA services. The first is to offer what she calls "multiple flavors" of DA, including call completion and charging a per-minute rate for DA, where a customer could call DA only once and get an infinite amount of numbers, being charged for time rather than having to keep calling the service back for more numbers.
Another way in which speech recognition is an attractive option for DA is as a low-cost way to offer what she calls "enhanced directory services," including movie listings, restaurant and traffic information or traveling directions.
Using live operators for such services currently is cost-prohibitive because DA providers have trouble monitoring and controlling how long operators stay on the phone with those calling in for the service.
"Third-party providers and the carriers that are offering these services are realizing they have almost no control over the operators and how long they stay on the call, yet they can only charge so much for these services," Frostad says. "It's very difficult to control operator time and many of them are finding it's not profitable. So speech recognition is ideal there, where a person could stay in the system as long as they want."
Liz Montalbano is news editor for PHONE+ magazine.
