Speech Recognition: Dialing Up Data The Easy Way

Speech-recognition services will soon offer a rich array of info over the phone

People don't seem to like the idea of putting on a headset and talking to their computers. That, as much as any technical shortcoming, is why speech recognition has failed to become a mainstream way of using computers. Tapping letters into a phone using the keypad seems equally odd. But talking into a phone is something that comes as naturally to most folks as breathing. That's why I think new services that use speech recognition over the phone are poised to become a major tool for getting information off the Internet.

I tried two: Tellme (www.tellme.com) and Quack (www.quack.com), both of which are still being tested. Other services, including BeVocal, Audiopoint, and TelSurf, are in various stages of startup. Although there are large differences in technology, content, and business model between Tellme and Quack, they have a lot of similarities. To use either, you dial a toll-free number, 1-800-737-8225 for Quack, 1-800-555-8355 for Tellme, but be sure to register on the Web site first. Simply speaking into the phone and listening to the responses enables you to get information such as traffic conditions, weather forecasts, stock quotes, and movie locations and showtimes.

In both cases, the speech recognition works well. A major source of errors in dictation software is the computer's poor ability to use context to understand what you mean to say and convert thousands of words into precise text. But phone-based systems need to understand only a few keywords. Once you say "traffic," they then know that any subsequent request concerns traffic conditions, not movies or stock quotes, vastly improving accuracy.

There are three technologies for getting information back out. Synthesized speech takes text and automatically but imperfectly converts it to spoken language. The result is full of odd inflections and can be hard to understand. Sampling assembles language from snippets of human speech, producing a sort of audio ransom note, with different words or phrases read by different voices. The most natural, but least flexible, approach is to select responses from recorded scripts. Both Tellme and Quack employ mixtures of these methods, but Tellme mostly uses canned recordings, while Quack uses a lot of sampling, which makes it harder to understand.

Overall, I found Tellme more appealing. The company is run by an unlikely mixture of onetime antagonists from Netscape Communications and Microsoft. Their goal is to build a voice-based information service supported by brief, relatively unobtrusive ads ("Tellme movies brought to you by New Line Cinema") and, ultimately, revenues from transactions, such as ticket sales.

SIMPLICITY. Tellme's test service offers traffic, weather, and restaurant and movie information for most major cities, plus news, stocks, horoscopes, and games. The current version requires you to register using a phone number and defaults to the location of that number as a starting point; later versions will rely on caller ID or location information from mobile phones.

The services are about as uncomplicated as you can get. Traffic and weather reports sound a lot like what you would hear on the radio. Restaurants are listed by types and are rated by "Tellme stars" and, where available, Zagat reviews. Stock quotes are available for any listed company, but the system had trouble understanding names and asked that I enter the ticker symbol from the keypad. My only real complaint is that the sound quality was often poor, like using a wireless phone with a weak signal.

Quack intends to make its business by providing phone-based information services to other Web sites. Its first deal, announced but not yet functioning, is with Lycos. Its stand-alone test service offers information similar to Tellme's, though restaurant lists are available only for Minneapolis-St. Paul and San Francisco. Its traffic reports tell about specific incidents on highways you ask about rather than the overall situation. For stocks, it allows you to set up a portfolio and get all the quotes with a single request. Sound quality is generally better than Tellme's.

These services are attractive because they use phones as they were meant to be used--for voice communications. They may have a hard time making money, and the field is already way too crowded. But I hope a couple of them make it, because this is a service I could really grow to like.

Before it's here, it's on the Bloomberg Terminal.