February 2001

TIPS & TRAPS

by Lowell Rapaport

IVR: A New Way to the Web

While there may be a digital divide, more than 95 percent of Americans have a telephone in their homes (not to mention the millions of mobile phones now in use). So why not voice enable the Web? Many companies are doing just that with voice recognition and interactive voice response technology.

You can voice enable your site by adding an interactive voice response (IVR) box. This is essentially a computer that performs two tasks. First, it takes content from a Web server and makes it available to callers by either playing prerecorded messages or by using text-to-speech software to convert content. The second task is to convert requests from the caller into data that can be understood by the Web server. This can be done by instructing the users to press numbers on the telephone keypad or by recognizing their spoken instructions using voice recognition software.

Business-to-business portals and corporate intranets are particularly good candidates for voice navigation. "Voice portals" can provide many of the same services as regular portals, such as storing personal data like addresses or appointments. Businesses can also make information such as customer records available to mobile employees.

In a customer service application, voice recognition technology takes the place of a receptionist or service representative. Users can help themselves to the information they need from a menu of choices.

According to Gerald Karam, division manager of innovative services research at AT&T Labs, VoiceXML will make it easier to deploy this technology.

"XML's ability to repurpose output makes it possible to voice-enable Web sites," Karam says. "VoiceXML lets voice servers match every change made to their Web counterparts. Without VoiceXML, voice servers would lag behind their Web counterparts due to the complexity and high cost of integrating voice recognition and text-to-speech technology with Web servers."

There are a number of applications that can't be easily handled using voice alone. For example, information extracted from a voice-enabled Web site can't be copied to other applications. Nor can you truly surf using voice alone.

"VoiceXML is more constrained in the sense that users are restricted to the information available on the [voice] server," Karam admits. In theory, multiple voice-accessible Web sites could link to each other, but since voice technology is not as widespread as HTML, there is little reason for the owners of voice sites to link together.

Perhaps the biggest limitation to using voice to navigate a Web site is the essentially serial nature of voice communication. Voice technology is limited to a single communication stream to the user. Another restriction is speech recognition technology. While the latest products are very good, they're still limited to prepared vocabularies and grammar. "It is much harder to do natural language recognition than directed response," says Brian Michael, senior product manager for BeVocal, a developer of voice portals and voice-based Web site infrastructure.

Directed response calls for the words to be spoken to fit in a limited range of expected responses. The best analogy to the world of paper is handwriting recognition, which is most often used to decipher expected values such as courtesy amounts on checks.

Nonetheless, VoiceXML and voice navigation will open up Web sites to many more users, reaching beyond the digital divide.

Cram Course: Application Servers

No doubt you have heard the buzz about application servers. In fact, they have been around for some time. They started out as middleware in three-tiered enterprise systems; the application layer placed between the client and the database. When the Web rose to prominence, so did application servers. It was natural to use them to go between users on Web browsers and databases.

An application server transfers data between users and databases and between applications. Essential features include a development environment that lets you create the applications in either a scripting language (Visual Basic, Perl) or a full-blown development language (C, Java). Enterprise application servers should also provide advanced features like load balancing, failover and caching to make the application more reliable and efficient.

Popular deployment environments and tools include Java 2 Enterprise Edition (J2EE), CORBA, COM and ActiveX. Application servers can be written for a specific platform, like Windows or Linux, but there is a growing trend to write the server in Java and run it from within a virtual machine. This gives the application server total portability.

Application servers vary in their intended application. Besides enterprise application servers (Oracle Transaction Server, Sybase Enterprise Server) there are electronic mail servers (Delano e-Business Interaction Suite), groupware or messaging servers (Lotus Domino) and electronic commerce application servers (IBM Websphere). Costs vary from tens of thousands per server to less than $100 (Ganjo). With so much variety, everyone should be able to find one that fits their needs.