Last year it looked as if speech recognition was finally ready for the mainstream PC. In rapid succession IBM, Dragon Systems, Lernout & Hauspie and Philips each rolled out what they claimed were highly accurate, low-cost programs that brought dictation and voice control to the desktop.

But today those companies are deemphasizing the PC sector as they drive toward the latest hot spot in voice recognition: network-based systems that use speech to access the Web or other information sources by means of phones and embedded devices.

IBM Corp. this year reorganized its Voice Systems group (West Palm Beach, Fla.) to focus on bringing speech recognition to Internet, embedded and telephony applications. Philips Electronics' Speech Processing unit (Dallas) shifted its attention from consumer dictation software for the PC to telephony and embedded systems. And Lernout & Hauspie Speech Products NV (Ieper, Belgium) bought Dragon Systems Inc. in March, in part to provide a path to new telephony applications.

"Dictation turned out to be more limited a market than companies thought and when the prices on the software dropped, they weren't generating as much revenue either," said William Meisel, president of TMA Associates (Tarzana, Calif.), a speech industry consulting group.

Instead, "There's been a major shift to network-based speech solutions based on the growth of wireless," said analyst Mark Plakias, vice president for voice and wireless commerce at The Kelsey Group (Princeton, N.J.).

Network-based speech applications, including voice portals to the Web, reach out to anyone with a phone, including a growing pool of cellular users looking for a way to tap into the Net. These technologies represent the intersection of wireless and the Internet, a phenomenon one analyst calls "the sweet spot for speech."

The market for voice portals could generate $1.6 billion annually by 2005, according to researchers at Cahners InStat Group (Scottsdale, Ariz.), while the overall market for network-based speech products is expected to reach $12 billion the same year.

Analysts like Meisel believe the PC sector flopped when consumers discovered they didn't feel comfortable composing their thoughts out loud, dictation-style. Also, he said, users were impatient with the time it took to train the programs to recognize their particular voices. Many came to see that — despite the high-tech chic — they just didn't need voice recognition on their PCs.

"Dictation was part of the first wave of speech recognition applications that occurred when PC growth was expanding and it seemed to promise the Star Trek effect," said John Meiling, vice president of global marketing for Philips Speech Processing. "But the reality is different. Most of the desktop dictation systems are speaker=dependent and take time to train, and some of the capabilities are limited." It turns out that with the PC, "it's easier to use the keyboard." For all the R&D that went into improving the accuracy of PC dictation software, it was a tough sell to consumers, Meiling added.

That's not to say speech recognition on the desktop is dead. IBM, Philips and L&H continue to sell their PC products, which some analysts said still represent the lion's share of voice recognition revenues, albeit the slowest-growing chunk. And PC software behemoth Microsoft Corp. expects to roll out its own long-anticipated voice recognition products for Windows early next year.

But all the companies now have potentially bigger fish to fry. One sign: the success on Wall Street of speech companies in the network solutions space, The Kelsey Group's Plakias said. Stock for both Nuance Communications Inc. (Menlo Park, Calif.) and Speechworks International Inc. (Boston) — network-based speech solutions vendors that went public this year — are trading between $90 and $100, roughly four to five times their offering price.

Nancy Jamison, a speech industry consultant who heads Jamison Consulting (Woodside, Calif.), believes crossing over into new markets "makes sense" for dictation companies because telephony application areas like call centers are using speech recognition to perform data mining.

The voice Internet

For its part, IBM revamped its speech recognition group to focus on what the company calls the "voice Internet," according to Nigel Beck, director of marketing for IBM's Voice Systems group. The move folded IBM's telephony group, focused on call center platforms, into the speech group and aggregated the operation with the company's Web server group to create the WebSphere Voice Server, which allows for the creation of voice-enabled Web applications.

Big Blue's roots in the speech business go back to the 1950s, and it always approached the technology from the computer side. But the company's new vision is to "leverage its strength in telephony and Web server platforms and combine those with its capabilities in speech," said Beck.

On the embedded side, IBM recently inked a deal with Canon to create speech-enabled consumer devices such as kitchen appliances, toys and game consoles. The companies have yet to hammer out the details, but products resulting from the collaboration could be released in as soon as six months. Separately, IBM forged a deal with Nokia to share research for speech-enabling with the mobile market.

IBM is also a founding member of the VoiceXML Forum, a group that is creating a standard for a voice-enabled markup language to help accelerate the development of Web voice content and services.

IBM's ViaVoice dictation software for the PC remains "one piece of the puzzle" of an end-to-end solution, Beck said, alongside PDAs, phones and the back-end servers that run the speech recognition and Web application infrastructure.

Talk to the TV

For its part, Philips Speech Processing expanded its capabilities last year by acquiring Voice Control Systems (Dallas) for $60 million. That company supplies a DSP-based network speech solution called SpeechWave. Philips sees the acquisition as helping to broaden its speech solutions in telephony as well as consolidating its North American-based speech research, engineering and marketing resources. Philips also has its own network-based solution, SpeechPearl, which resides on a host server.

Philips is less bullish than IBM on bringing voice recognition to the Web, but it does see a role for the technology in embedded systems. The company has focused on developing chip-based speech solutions in concert with Philips Semiconductors for its $12 billion consumer electronics business, and now has prototypes of speech-enabled TVs, DVD players and a mobile handset. The Philips-brand Command TV will hit the market at the end of 2001.

Similarly, embedded systems are also a target for Lernout & Hauspie. Its Dragon Systems unit has introduced a handheld digital recorder that runs speech recognition software for doing e-mail and PIM entries. And L&H has licensed its technologies to mobile-software developer Symbian for use in smart phones. L&H also sells its RealSpeak text-to-speech technology to telephony developers.

Consumer dictation software sales make up less than 10 percent of annual revenues for L&H. Revenues from sales of professional-quality dictation for medical and legal markets are higher. The March acquisition of Dragon was aimed at increasing L&H's share in the dictation market, and also to provide a pathway into telephony, where Dragon has been more successful.

The acquisition added 170 researchers and engineers from Dragon Systems, but it's not yet clear what the fate of the Dragon researchers will be. The company has been mum about its plans since the launch of a U.S. Securities and Exchange Commission investigation into L&H accounting practices.

Indeed, the once high-flying L&H is going through a period of gut-wrenching change. Its two founders — Jo Lernout and Pol Hauspie — announced last week they would give up their jobs as executive co-chairmen, ceding the role to Roel Pieper, a former senior Philips executive. In addition, L&H said it would have to delay its upcoming quarterly report and restate its finances for 1998, 1999 and the first half of 2000 due to irregularities uncovered in an ongoing audit.