02/23/19 ONLINE ORIGINAL: TALKING YOUR WAY THROUGH THE WEB

Return to main story

TALKING YOUR WAY THROUGH THE WEB

He's like a cross between a square-dance caller and the smooth-talking professor in Apple's famous Knowledge Navigator video. Seated in front of a notebook PC, microphone in hand, Ron Risdon barks out a string of instructions and watches the computer jump from one topic to another through that vast information network, the World Wide Web.
Risdon, executive director for business development at Redmond (Wash.)-based Conversa (pronounced "Conver-say") Corp., is talking his way around the Internet using his company's new voice-controlled browser, Conversa Web. The software, set to be launched on Feb. 16, builds on Microsoft's Internet Explorer 4.0 and requires no special hardware other than a microphone and a standard sound card. (Conversa recommends a 100-MHz or faster Pentium PC configured with at least 24 megabytes of RAM. The software runs on Windows 95 or Windows NT.) It will sell for $29.95 over the Internet or $39.95 at retail outlets.
Conversa Web looks simple on the surface, but behind the scenes are years of research into linguistics and speech recognition. The software knows roughly 20 stock phrases that control movement around a Web site, such as "scroll down" and "bottom of page." But the real trick is that Conversa Web can navigate among links without having been previously trained in how to recognize them. The secret: A powerful phonetic engine in the software scans each Web page to find links, then predicts how they'll be pronounced, and then listens for the user to say them.
That's what sets Conversa Web apart from IBM's VoiceType Connection, a speech extension to Netscape Navigator 3.0 that came out last year. IBM's product, like Conversa Web, was speaker-independent and able to navigate the Web. But it couldn't follow most links unless it had been trained how to recognize them. That limited its usefulness for unconstrained surfing.
HEY Y'ALL. In a demonstration of Conversa Web for Business Week, Risdon and a reporter were both able to surf the Web hands-free, even when the reporter laid on heavy Southern, English, Indian, and French accents. When Conversa Web talked back--asking for more guidance or prompting user input--it used high-quality digitized speech, available in either a male or female voice.
The product has some drawbacks, though. It uses a customized version of IE 4.0 that doesn't support all its features--in particular, the Channels function used for "Webcasting" information to a PC that can then be viewed offline. Worse, Conversa Web can't understand or perform common Windows 95 tasks, so anything other than browsing has to be done using the keyboard and/or mouse. It's a bit disconcerting to be able to talk your way through the Web and then have to grab the mouse to do something as simple as minimizing a window. Risdon says the software's next release might add basic Windows "command and control" functions if customers request it. This could be especially important to disabled users, who'll find Conversa Web a powerful tool.
The privately held Conversa has an intriguing history. CEO Steve Rondell is a linguist who in the late '80s launched a handheld computer, called the Voice Computer, that had voice input/output and could translate between different languages. Far ahead of its time, the product failed because a key chip supplier pulled out. Rondell formed Conversa in 1994. The company is privately financed by 65 investors, including two actors from Star Trek TV shows: James Doohan, who played Scotty in the original series, and Brent Spiner, who played Data in Next Generation. Conversa is located in Redmond because Rondell lives there, not because of any ties to Microsoft.
Risdon joined Conversa last September, when it merged with his company, Medius, which was working on a unified messaging software package. That union formed the basis for Conversa's next product, a voice-enabled messaging package called Conversa Messenger. Due in April, Messenger is a voice-based messaging application that runs on a PC. Think of it like Microsoft Exchange enhanced with speech--or like Wildfire Communications' Wildfire big-ticket message processing system, only client-based instead of server-based.
CALL MOM, PLEASE. Conversa Messenger lets users dial up their PCs and interact with voice-mail and E-mail messages using spoken commands. For instance, users can ask to have voice-mail messages played to them and then command the software to patch through a callback. Messenger can also read incoming E-mail messages aloud (using robot-like speech synthesis) and accept digitized spoken replies that are returned to the sender via E-mail. (It doesn't allow users to dictate text messages, however.) Voice and E-mail messages can be saved, forwarded, or deleted, just as if the user were sitting in front of a PC.
The drawback to Messenger, as with Conversa Web, is that the software isn't merely an add-on to a popular off-the-shelf package. Rather, Messenger uses its own in-box, which could frustrate users who would prefer to use a mainstream product like Microsoft Exchange. (Messenger's in-box does follow industry standards, however, so it can be accessed from other mail packages.) Messenger will sell for $149.95 at retail stores or $99.95 over the Web.
Conversa Web and Messenger typify the current state of the art in speech recognition--with all its progress and shortcomings. Though the company says its packages use "continuous," or natural-language, as opposed to "discrete," speech recognition, the reality is that they recognize only a limited number of untrained phrases (things like "scroll up" or "go back"), all of which are essentially handled as one long discrete word. Conversa doesn't understand commands that it hasn't been programmed to respond to--with the exception of the sophisticated parser that analyzes Web pages to figure out how the links will be pronounced. If the software is given a limited universe of possible phrases to recognize, it does a great job.
But for all this sophistication, we're not yet at the point where you can say to your computer "Tell me about the rainforest" and have it deliver the knowledge you seek. The Knowledge Navigator is still a pipe dream.
By Andy Reinhardt in San Mateo, Calif.

Return to main story