By Margaret Young
Remember what 2001 was going to be like? Space stations on the moon, men on Mars, and computers that could talk. Well, lunar vacations won't be here anytime soon, but computers that can hold a conversation are a long-awaited fantasy whose time is nearly here.
Nuance Communications Inc. of Menlo Park, Calif., is leading the virtual gabfest. A six-year-old spin-off from SRI International, formerly the Stanford Research Institute, Nuance uses a combination of algorithms, math, statistics, acoustics, and linguistics to create software that enables computers to comprehend and respond to human speech. Such voice-recognition software, enthusiasts say, could radically change customer service by having computers handle routine calls and transform the Web by making an audio version of it accessible to anyone with a phone.
In an industry that has had numerous false starts, Nuance is starting to bear fruit by developing server applications that customers can reach by phone, using simple speech to direct the computer. Others have focused on having desktops take dictation, which has been more problematic.
A BILLION PHONES.
Today, Nuance's system can be found in the customer-service departments of plenty of the largest corporations, including Fidelity, American Airlines, Verizon Communications, and Charles Schwab, which in 1996 introduced its Nuance-based "Schwab by Phone." After dialing an 800 number, customers receive a menu of options similar to that of a touch-tone service. But instead of punching in requests, the customer might get a stock quote by speaking the company's name. "Unless you're an active trader, you don't know the [stock] symbols" you need to type on the telephone keypad, says Steven Soso, Schwab's vice-president for new-product development. "For our customers, [the voice system] was like talking to a rep" or broker.
Currently, the voice broker handles about one-third of the 56 million calls Schwab deals with automatically every year. More recently, Schwab introduced voice-automated services that allow computers to handle routine inquiries on account balances and other activities previously handled by service reps or the customer navigating a cumbersome touch-tone menu.
The real growth for voice recognition, though, is likely to be in the so-called voice Web -- an audio version of the Internet that is reached via a voice portal, or voice-automated system that leads the caller through audio versions of pull-down menus. One such portal is tellme.com, which uses Nuance. Dial 1-800-555-TELL, and you can choose among categories such as news, finance, and entertainment. Say the title of a movie, and the site will tell you where it's playing in your neighborhood, at what times, and offer an audio review.
Nuance CEO Ronald Croen points out that while there are 300 million PCs connected to the Internet, there's already an installed worldwide base of 1 billion phones, giving anyone easy access to the Web (Video Interview, 2/26/01). Kelsey Group, a Princeton (N.J.) communications-industry consultancy projects that the voice Web will be a $12 billion market with 45 million customers by 2005, up from virtually nothing now.
"NOT A PERFECT SCIENCE."
But Nuance and others still need to do some fine-tuning before voice-recognition software becomes the talk of every town. While having an open-ended conversation with a computer is possible in laboratory situations, the commercial use of the software tends to be limited to directed dialogue, in which the customer responds to a series of instructions and questions like: "State the name of the person you are trying to reach." The technology for natural-language recognition requires too much processing power and bandwidth for large-scale commercial applications. In addition, though voice-recognition software tends to work well in a quiet office, many users complain of problems when they try to access a voice portal from a car or busy street. "Speech recognition is not a perfect science," says Marci Gottlieb, a spokeswoman for Tellme Networks, a voice portal that uses a Nuance platform. "If you're in a loud environment you run into trouble."
As it works to stamp out the bugs, Nuance is also seeing its sales grow. The company reported revenue of $51.8 million for 2000 -- a 165% increase over the previous year. Although it still lost $17.4 million, it could reach profitability by the end of 2002, says Bill Hills, an analyst at Aberdeen Group. Nuance's performance was still better than its closest competitor, Speechworks International Inc. in Boston. Speechworks posted $30.3 million in sales and a net loss of $36.6 million during the same period. "Nuance is clearly one of the leaders in providing the basic technology that makes this all possible," says William Meisel, president of TMA Associates, a speech-industry consulting firm in Tarzana, Calif.
Croen couldn't agree more. "No one is really asking anymore, 'Does speech recognition really work?'" he says, noting his customers handle hundreds of thousands of calls a day. "The proof of that is already out there."
And it's only a phone call away.
Margaret Young is a freelance writer based in San Mateo, Calif.