Voice recognition meets the Web: New technology targets vertical markets

When E*Trade Group Inc. first made its appearance on the Web, it triggered a ripple effect in the securities industry that caused many of Wall Street's biggest names to jump into cyberspace.

Now, the company that made its name allowing small investors to trade stocks and options over the Internet is attempting to do the same thing with voice recognition. Using a system from Applied Language Technologies, E*Trade has added an option to its Tele*Master system that lets callers check quotes and place trades by speaking their commands. Although E*Trade is not the first brokerage company to use speech recognition-Charles Schwab has a similar system-its application is the only one that completes transactions without independent verification through touch-tone.

Behind the system sits ALTech's SpeechWorks system, which consists of several modules that can be interchanged depending on the application. And not surprisingly, many of those applications are getting their start on the Web, according to Stuart Patterson, CEO of ALTech. "There's no reason speech recognition can't enable a whole bunch of services the same way the Web does."

In fact, the E*Trade application is just one example of what many speech recognition companies are expecting to be a flood of new applications that tie their systems together with existing Web databases (see figure).

"We're still in investigation mode, but there's a lot of interest in deploying a Web interface and speech recognition systems," said Steve Erlich, vice president of marketing for Nuance Communications.

Ironically, though, vendors are finding much of their recent successes in more mundane applications such as basic customer service. Among ALTech's customers are BellSouth and AT&T, both of which are using voice recognition to enhance existing products.

In BellSouth's case, the carrier has added voice recognition to its talking Yellow Pages, allowing callers to speak the name or type of business for which they are looking.

"Speech recognition is a way of differentiating themselves even when the application is something that's almost a commodity," said Patterson.

It's just those applications that also are giving the technology a boost, although they're not the only factors.

Perhaps the biggest driver behind renewed interest in speech recognition is the improvement in accuracy, particularly on voice-independent systems, which recognize vocabulary from any speaker. Users no longer need to tolerate "training sessions" during which they must speak commands to a system while software measures voice parameters. Also gone are the limited vocabularies that restricted users to a small set of commands.

"On speaker identification, you get such high recognition that there's no real reason to tune your voice," said Patterson. "Speaker-dependent technology" in which applications are tuned to a specific voice, "is an added level of complexity that a lot of applications don't need," he said.

Indeed, accuracy on voice recognition has become so good that it's often not a major factor in application development, said Patterson.

"Accuracy is not as important as transaction completion. As long as you can smoothly correct mistakes, the customer tends to have a good experience."

Also contributing to the technology's rise is the fact that companies are building systems on open platforms. For most, that means building tool kits that allow third parties to develop applications in well-known languages on accepted platforms.

"The problem with the telephony business in general is that it's fairly proprietary," said Erlich. "We're trying to bring the open environment to that market."

Nuance's latest software tool kit, which is in beta tests, includes support for Windows NT and Digital Unix.

In addition, the company is supporting Active X, which Erlich said should open up the developer market to a wide audience.

"It's an environment that these developers know a lot about. And secondly it drastically improves the time to market," he said.