Computer, status report. (voice recognition software) (Multimedia PC) (Column)
by David English
When you hear Captain Picard casually call up the ship's computer on "Star Trek: The Next Generation," do you ever wonder when you'll be able to control your own computer by voice? We're a long way from having intelligent computers, but simple voice recognition software (also called speech recognition software) has finally arrived. By the time you read this, many of the popular sound cards will be shipping with some form of voice control software. In this column, I'll discuss how well this software works, what's available, and who might want to use this futuristic technology.
A number of factors determine how accurate a voice recognition program will be. These include how well the software is designed, your consistency in pronouncing the words, the number of commands that are active (especially similar-sounding words), the quality of your microphone and sound card, and the level of background noise.
Most voice programs are speaker-dependent, which means they work best when you train them to your own voice, repeating each command three or four times during an initial training session. Most programs let you save multiple command sets so that more than one person can use the program.
With a good 8- or 16-bit sound card, decent microphone, relatively quiet environment, and careful training, most of these programs are between 80 and 95 percent accurate. That means they'll only occasionally confuse one command with another or require you to repeat a command a second or third time.
How does a voice program work? Most programs record your voice command and compare it to a database of trained commands using a pattern matching algorithm. The program calculates a score that represents how close your voice command is to each trained command and chooses the trained command with the closest score. If the score is within the acceptable limits, the program initiates the macro that's associated with the trained command. If the score is outside the acceptable limits, the program alerts the user that it didn't understand the command. The entire process takes place almost instantly.
The first major sound card to ship with voice software was Microsoft's Windows Sound System. Its software, called Voice Pilot, includes command templates for many Windows applications. It also lets you add your own commands. Media Vision's software, called ExecuVoice, is nearly identical to Microsoft's Voice Pilot because Dragon Systems wrote both programs. Media Vision uses a newer version of the Dragon engine, but it doesn't include the ability to add additional commands (though you can launch any new application). Both ExecuVoice and a button style microphone are included with the new Pro Audio Studio 16 sound card and in an upgrade kit for the Pro Audiospectrum 16. Dragon also offers two voice programs through IBM. They're called IBM VoiceType Control for Windows ($129, with microphone) and IBM Voicetype 2 ($2,195, with a vocabulary base of 7000 words).
The voice software that Creative Labs is bundling with its sound cards was developed by Voice Processing of Cambridge, Massachusetts. it's called VoiceAssist, it supports as many as 1024 commands per application (256 active at a time), and it adds a built-in macro program. VoiceAssist is shipping with all Sound Blaster 16 sound cards. Creative Labs offers an upgrade path for current Sound Blaster 16 owners.
Covox has been making voice recognition products for many years. In addition to its own sound boards with voice support, Covox offers a program, called Voice Blaster ($119.95), for Sound Blaster-compatible sound cards. It includes both DOS and Windows interfaces, support for Windows OLE, and a headset with a microphone and earphone. Digital Soup plans to release a basic voice program called Rover ($129, with an introductory price of $49) that translates voice commands into keyboard commands. And Sierra Semiconductor is providing a number of sound card manufacturers with its new hardware-based Aria Listener technology. Some of these boards are shipping with a Star Trek game from Interplay that can be voice controlled.
Who would want to use voice recognition technology? It's most appropriate for handicapped individuals, though a less-than-perfect recognition rate could cause the computer to hang occasionally. It's also appropriate for special work situations, where an employee must use his or her hands doing something else, as on a factory assembly line. For the rest of us, voice programs provide a great way to impress our friends and relatives, though I suspect few of us will use voice control every day.
Perhaps in five years or so, when the technology is perfected, you'll be able to walk down an office hall and hear people say, "Print page three" or "Send memo to George Smith." We may also have to contend with the disgruntled employee who yells, "Reformat hard drive!" as he's escorted out the front door.