Beyond Voice Recognition

Imagine a world where we no longer talk to each other but instead have conversations with machines that we have programmed to have personalities -- good ones, of course. We, in our infinite wisdom, sit in our cushy La-Z-Boys and dictate instructions for the machines to carry out. While we recline and sip champagne, the computers talk to each other, accomplish the tasks, and then report back to us -- in our language, of course.

The concept is far-fetched, perhaps impossible even, but we have automated our lives to such an extent that we are becoming increasingly efficient. Much of that efficiency stems from wireless service offerings.

Adding speech recognition technology to existing wireless services as a replacement for keypad punching is a natural progression. As the number of on-the-go subscribers increases, so too does the necessity to make the task easier and safer. In the last few years, enhanced services have come about as tools that wireless carriers can offer their customers to differentiate themselves in the marketplace and to help their customers become more efficient. Vendors are taking these services and enhancing them even further with next-generation voice-recognition technology. Now carriers can offer their customers efficiency with a twist.

FROM WHENCE IT CAME Speech-recognition technology has evolved over the past 10 years from government-funded university research, said Stuart Patterson, Applied Technologies (ALTech) president. Then many companies began working with the technology and sharing their work with the universities.

"We ended up taking technology out of MIT three years ago to found our company," Patterson said. He explained that all of the companies used the same basic approach, which was to break up the voice signal and use efficient algorithms to match one speech wave against a database of speech waves.

"They try to figure out against a huge database of wave patterns how close the match is. Then they come up with a confidence score that says, 'Oh, it looks like it's actually this one.'" Patterson said. Recognizing a word, such as a roaming access code, or a name or the way someone says his name is not done in a fundamentally different way.

"It's done with essentially the same core technology," he said.

Companies have taken that core technology and gone after different niches in the market. Trying to separate the niches is like trying to slice a Jell-O mold. The lines all blur so that it's difficult to distinguish the sections. If you look closely enough, however, you can detect the lines. Companies are specializing in speaker verification, enhanced services such as personal assistants and single-number services, and CPE, or call- center-type services.

APPLICATIONS Wireless carriers are most interested in enhanced services and speaker authentication. AccessLine has updated its personal number system with ALTech's speech recognition technology. Now carriers that offer customers AccessLine's 1-number service can take that one step further and offer them the speech- recognition interface as well.

"We've developed a voice-activated user interface that does the same thing (as the Touch-Tone version) but with voice," said Kimberly Tassin, AccessLine Technologies corporate communications manager. So instead of pressing "1" to listen to messages, the user says, "listen to messages." The system will let you know how many messages are waiting and ask you what you want to do with each one. The user can say "save," "delete," "next message" or even "direct my calls to my home for two hours." All of the tasks that could be performed before with the keypad now can be performed by voice commands.

"We worked nearly a year and a half with ALTech to develop all the language for the application," Tassin said. "The reason we chose ALTech is they have certain capabilities that we thought were very important for a robust system. It's based on phonemes, so you don't have to train the system on your voice like you do with a lot of other voice technologies."

The system is network-based, so when you purchase the system, it sits on the network and is attached to the central office, or the mobile switch. It is a self-contained unit, consisting of a control computer, a switching matrix and all of the AccessLine modules.

"When a call comes into the carrier's network and is recognized as an AccessLine number, the call gets sent to the AccessLine system, which processes it and sends it out where it needs to go," Tassin said.

Wildfire provides a similar application. The company incorporated voice-recognition technology into its Wildfire Personal Assistant from the beginning, and instead of partnering with a company that specializes in speech recognition, the company opted to handle all of the work in-house.

According to Leslie Anderson, director of corporate communications, the original version was designed as a CPE system, a customer premise equipment box that was sold on the voice-mail model. About a year ago, however, Wildfire brought on a new CEO, Robert Mechaley, formerly of McCaw and AT&T Wireless, who changed the company's direction to focus on the wireless segment of the telecommunications industry.

"We basically took the existing technology we had and ripped out the speech recognition. It was a third-party board at the time and was a real barrier to having the system scale up to tens of thousands of subscribers, which is what a carrier would need," Anderson said. "We wrote our own recognizer all in software so it could exist in an entry-level, inexpensive Pentium computer and live independent of the carrier switch. It can just sit in the network, and the recognizer is really ruggedized for the kind of noise that people encounter in the wireless environment."

One aspect that the company touts is the human-like qualities of the personal agent.

"Wildfire is really an intelligent entity," Anderson said. "You really think of her as a colleague. You form a relationship with her." Anderson said that although the company offers wallet cards for carriers to distribute to customers, carriers don't need them because Wildfire will walk their customers through each step. She said that PacBell, in its trial, chose not to use the cards, wanting instead to see how people would do with the on-line hinting and prompts that are in the product.

"Our ideal is that customers start out very simply," Anderson said. "It's a consumer product, and we don't want users to be overwhelmed with the features, which is probably the biggest difference between this version of Wildfire and the version that was out there."

The carrier version is modular. It starts out with voice dialing and voice-activated voice mail. After the customer has completed a carrier-determined number of sessions, Wildfire leaves a message offering the user another feature. With each successive feature, Wildfire offers the customer a tutorial, and Wildfire self-provisions the account, so the user does not have to contact customer-care personnel. "We have basically broken it down into a Chinese menu, so it's a lot of little bite-size pieces that the carrier can configure," Anderson said.

Wildfire sells the carrier the software, which runs on a Compaq Proliant server platform. Then the company loads the software onto the server and integrates it into the carrier network.

Another vendor that has done a lot of work with speech-recognition technology, albeit on the wireline side, is Periphonics. Vinay Chandhok, director of systems engineering for major accounts, said that because the wireless industry is evolving now, it could avoid the mistake the wireline side initially made with the old voice-mail system.

"If you remember, we started off with (answering machines) in the house and as time went on, the wireline industry realized that there was good business and certain advantages to the customer if you put voice mail on the network," Chandhok said. "But we were fighting an embedded base of CPE equipment and trying to justify how the network would be better." Chandhok said he sees a possible parallel in the wireless industry with companies integrating speech-recognition technology into the handset.

"I think there will end up being a limitation on what you can do in the future through the network, and again we are going to kind of have things in the reverse order," Chandhok said. Periphonics also is doing a lot of work with voice- authentication technology.

"There are several services where privacy is involved," Chandhok said. "You wouldn't want somebody calling with your handset and picking up your messages." The handset should be able to tell by the accent or tone if the subscriber is not the one using it, and then deny the user the menu that normally would be available, he said.

"We have a couple of pilots going, and we are working pretty feverishly toward getting (voice authentication) on the market," Chandhok said. "I think in the next couple of years, you'll see some real services coming on."

IMPROVEMENTS Some enhanced services that use voice-recognition technology, such as voice dialing and voice-activated voice mail, have been around for a few years. It is only recently that carriers have been able to offer their customers interaction with an electronic "person" who transfers calls, routes faxes and keeps track of their whole phone directories. Because of advancements in the core voice-recognition technology, carriers now have the ability to expand and tweak their offerings.

"Speech recognition has really reached a turning point in the past two years," Patterson said. Many companies, as well as universities, are continually working on improving the algorithms. But the big reason, Patterson said, is that processing power has been doubling every year and a half for the same amount of money.

"Now we can throw much more processing power for many fewer dollars at the problem, which is basically the database-search problem," said Patterson.

Patterson added that another sign of maturity he sees in voice-recognition technology is in the tools used to deploy it and the knowledge of how to structure the user interface.

"We believe that the speech-recognition accuracy is only 20% to 25% of the challenge," Patterson said. The rest of it is how you integrate the technology in a system that is scaleable, robust and deployable, and how you structure the application so users will accept it and come back again and again, he said.

"And since you know it's not going to be perfect 100% of the time, one issue is handling an error in a nice way," Patterson said.

Tassin pointed out that there always will be times when users don't want to speak out loud, perhaps because they are in a busy airport with lots of background noise, or perhaps they are in the middle of a meeting. With AccessLine, users always can override with Touch-Tone, and in most systems, users can bail out to a human operator if necessary.

WHAT'S AHEAD There's no rest for the weary. Vendors continually are looking to the future, striving to come up with the most intriguing and useful products and services for carriers to offer as well as to improve existing ones.

"Let your fingers do the walking" may become "Let your voice do the asking." BellSouth IntelliVentures, the electronic publishing arm of BellSouth, which provides Yellow Pages information in a variety of electronic forms, is conducting a technical trial in Daytona Beach and Gainesville, FL. The company is offering a Yellow Pages-type service using voice-recognition technology. David Shipps, new product manager, said the company would like to offer a similar service to wireless users.

In the trial markets, the service works like this: A user dials the service and says he is looking for a Chinese restaurant in Gainesville, which kicks off a query so that the service fills out an electronic form that helps it search the database. The service then will ask questions, such as "What area of town are you interested in?" or "What price range are you looking for?" Once the system has narrowed the query, it will begin reading off the restaurants. Although the service does not give directions, Shipps said it will tell you the address and phone number and can even direct-connect you.

"Internally we've been shopping the idea to our cellular folks here at BellSouth," Shipps said. Although there was definitely some interest, they are still a way from launching the service. "They are obviously interested in providing new and unique services that are going to help their customers, and we're interested in helping them. It's just a matter of trying to figure out what the issues are. We've never tried this in the wireless environment," he said.

As far as improvements to existing technology go, Patterson said the future holds more integration of speech recognition into packaged applications and into application-developed environments. He said that software providers fall into two categories: ones that provide packaged applications such as voice mail and enhanced services platform providers that sell a development tool kit, also called a service-creation environment, which carriers can use to develop almost any application.

"Speech recognition needs to get into both of those as more of an integrated element," Patterson said. He said the future also holds improvements in the tuning process and recognition accuracy.

According to Tassin, a goal for the future is to improve the language capabilities to incorporate everything people might say.

"There are a lot of ways a person might say 'yes,'" she said. "They could say 'yeah,' 'uh huh,' 'sure,' or even 'go ahead.' Right now you have to say 'yes' or 'no,' and the system tells you that."

"The last thing is that we just hope that people in the near future will come to expect these systems," Patterson said. "Users are still not familiar with this possibility, and they need to become comfortable with it."

A SURE THING? Although vendors are dangling several voice-recognition options in front of carriers, they are not biting yet.

"There's a lot of interest in voice activation, which is why we developed it, but we haven't yet gotten to the point where a carrier says that they want to buy it," Tassin said. "It's an extra component that you have to buy to make the speech recognition run on the AccessLine system. I'm not sure what they're waiting for. Maybe they're waiting for someone else to try it first."

Wildfire doesn't have any customers yet for its network service either, although PacBell is participating in a trial.

Despite the slow start, however, industry analysts say that enhanced services are going to pick up in the future.

According to Mark Lowenstein, The Yankee Group vice president and director of wireless/mobile communications, enhanced services will become a key method for carriers to differentiate themselves from their competition as wireless prices continue to fall. Yankee's recent report, Enhanced Wireless Services: It's a Brave New World, states that changes are taking place in the wireless industry that will force carriers to deliver more enhanced services.

If enhanced services do become more prevalent, then enhanced voice-recognition services would move right along as well.

"What would you rather do?" asked David Berndt, program manager with The Yankee Group. Stop your car, get out your directory, look up the number and punch in the keys? Or, while you are driving, just say "Call John Brown" and the phone looks up the number in your personal directory and dials for you? Which is more convenient?

We tend to spend more time doing what comes naturally to us, and since most of us love to talk, it would follow, then, that we eventually would demand that speech-recognition technology be added to our wireless applications.

"I think if you can imagine any of these services, whether it's travel reservations or BellSouth's information service, all of those are going to be accessible to cellular services in addition to the standard sort of cellular applications such as personal agents," Patterson said. "So the number of applications is mind-boggling. If we can make them available via wireline phones, we can make them available via wireless phones."