In a few years, people will be able to order pizzas over the phone without ever talking to a person, change a TV channel with a word or two and file expense reports from their cars via cell phone.

Technology that allows computers to understand speech has been in the works for more than 30 years, but getting it right and widely accepted has proven more difficult than researchers expected.

Experts say they've finally worked through some major stumbling blocks such as unreliable cell phone connections and background noise that make it tough for computers to pick up a voice. The technology will likely go mainstream within five years.

They expect cell phones and cars to be the areas where voice-recognition could have the biggest growth. Concern over distracted driving makes using speech to dial phones and handle other tasks in cars even more urgent.

Some of the new technology, reminiscent of HAL, the talking computer in the movie "2001: A Space Odyssey," is already here:

- The Infiniti Q45 from Nissan allows passengers to change the temperature and audio controls with voice commands. OnStar, a Troy, Mich.-based subsidiary of General Motors, has launched a system in which subscribers pay a fee so they can make phone calls and get weather, stock quotes and e-mail read to them through computers that understand voice commands.

- United Airlines has a system that allows callers to get flight information by saying a flight number, or arrival and destination cities for a flight. The computer asks questions and the caller responds by voice.

- Sprint PCS recently launched a system that allows its customers to buy plans that will let them get basic traffic reports or horoscopes read to them by uttering commands into their phone's speakers.

But speech recognition technology is expected to become more sophisticated. Companies are spending millions to make that happen. The 185 or so worldwide companies working on speech recognition hope to cash in on what could be a vibrant, new market.

Growing volume

Forrester Research estimates that voice-based commerce will reach or exceed $450 billion, or three times the projected amount for online retail sales, by 2003.

"It's going to be very big," said Ira Brodsky, president of Datacomm Research Co. of Chesterfield, Mo. "Anybody who has a phone will be using these kinds of services whether they know it or not."

Brodsky said the technology will become popular because it will focus on practical uses like getting directions or traffic reports while driving.

The technology has improved, making speech recognition possible.

The cost of hardware and software has gone down. Faster microprocessors handle information in seconds versus the several minutes it took a few years ago. Wireless networks are more reliable.

"It all depends on having the infrastructure in place," said Judith Markowitz of Markowitz Consultants in Chicago.

Not quite conversational

Software that understands conversational speech has yet to be fully developed.

Emmett Coin, CEO of ejTalk Research, a speech recognition research and consulting company in Detroit, said speech recognition systems will need to understand conversation to be fully accepted by the average person. "The more it mimics our human innate pattern of speech, the easier it will be and the less we will know it is there," he said.

Most commercial systems require specific voice commands or work by recognizing particular words that act as commands.

For instance JustTalk Inc., an Ann Arbor, Mich., speech recognition company, has developed software that allows businesspeople on the go to update their computer calendars, contact lists and even their sales inventories with a phone call. While the system can't understand conversation, it picks up particular words from a caller's sentence.

Commands essentially limit the field of words a system has to recognize helping to increase the rate at which computers can accurately understand speech.

Much of today's software can reach up to 90 percent accuracy, even higher with training, experts say. But strong accents, high-pitched voices and speech impediments can throw accuracy rates off.

Background noise can also confuse voice-recognition systems. Unlike people, computers cannot focus on a particular speaker and drown out the rest of the background noise, explains Raymond Gunn, CEO of Clarity, a Troy company that makes software that enables computers to separate a speaker's voice from other noise.

Gunn said Clarity's software can boost a system's accuracy rate from 90 percent up to as high as 97 percent.

Background noise is a particularly big issue in cars and public places.