More MS news articles for June 2000

Straight Talk on Voice Recognition Software

By Mark E. Smith

Here I sit at my cluttered, coffee-stained desk, dictating an article -- which, appropriately enough, is about voice recognition software. Although I've logged over 100 hours using the leading dictation programs, I must confess that it's still strange to me, a bit like flirting with HAL 9000, the computer with an attitude in the movie "2001: A Space Odyssey." Yet this snippet of science fiction on my desktop allows me to do all but hit warp speed with my writing, going from 15 words per minute using a keyboard to 40 wearing a headset.

Sure, voice recognition has its quirks -- like not recognizing the word quirks -- but once you get beyond the idiosyncracies, you may find yourself among the thousands of us with disabilities liberating ourselves with inexpensive, off-the-shelf voice recognition software.

Talk is Cheap

Voice recognition programs first appeared in the late 1980s and were immediately marketed toward those with disabilities. After all, no one benefits more from voice-activated computer operation than those of us with limited use of our hands. Unfortunately PCs didn't have the power they have today, so in order to run the software, the user had to purchase a cutting-edge computer costing close to $10,000.

But even after spending 10 grand, you were using software that was marginal, at best, and you spent months getting the computer to recognize your speech. In order to build a vocabulary within the system, you had to read from a dictionary, inputting one word at a time.

But just like other computer technology, voice recognition software has evolved over the past decade, and today you can run voice recognition software on your PC for well under $200.

You still need hardware that can handle the demands of voice recognition programs, but it's affordable now. The manufacturer-suggested minimum system requirements are as low as a 133-MHz processor and 32MB of RAM. But the software runs painfully slow when used on minimal systems, causing on-screen results to lag behind your dictation. For efficiency, you should run a minimum of a 350-MHz processor, 64MB of RAM and 310MB of hard disk space. Additionally, you need a CD-ROM drive and microphone input capability. Most of today's computers easily meet practical system requirements to run all brands of voice recognition software, but if your computer is a few years old, you may need to upgrade your hardware.

Simply Speaking

While voice recognition programs come with built-in vocabularies, you must train them to recognize your speech patterns. You may love your Southern drawl or your Chicago accent, but they drive the software crazy, making it struggle to match your speech to its template. It no longer takes months, but unless you're a network newscaster, it is a tedious task.

Programs boast initial training sessions of under 30 minutes. It provides you a selection of on-screen texts to read aloud while the computer tunes itself to your speech patterns. The texts range from "The Wizard of Oz" to "3001: The Final Odyssey" to a list of famous quotations. I couldn't get through the cliché quotations without giggling and upsetting the machine. Most of us read aloud in a different manner from our natural speech, so when you do the initial training, it's important to read in your normal voice. The first training session produces only about a 60 percent accuracy rate. Most people have to repeat the session and continue training individual words over the long haul. A minimum of 15 hours is needed to get acceptable accuracy.

Most users report that when using the software, they must enunciate more than in everyday speech and have difficulty getting used to saying punctuation and commands such as "backspace" and "capital." Despite best efforts, even the most experienced users do not get perfect recognition, topping out at 95 percent accuracy.

The more you use voice recognition software, the easier it gets. And don't discount the lighter sides of the training process, such as when you dictate, "Our waiter was rude," and the computer writes, "Our waiter was nude."

Working It to Work for You

Technology is sensitive to human inconsistency, and voice recognition software is no exception. In order to maintain program accuracy, every aspect of you and your hardware must remain constant.

Here are some quick tips.

Users who experience fatigue may have difficulty with the software, since their speech patterns may not remain consistent throughout the day. In this case, it's wise to train and use the software at the same time each day.

Correct microphone placement is very important. If it's even a bit out of alignment from where it was during training, accuracy will be greatly reduced. As a rule, the microphone should be placed a thumb's thickness from the corner of your mouth. If you experience difficulty aligning the headset, a high-quality, desktop microphone might help, but also know that results are best with a microphone close to the mouth.

Most users still use the keyboard while dictating, correcting errors by hand rather than talking the system through corrections. So you're still tethered to the keyboard, but with the voice recognition software, you are saving 80 percent or more of your keystrokes.

Hot Programs

Dragon Naturally Speaking Preferred 4.0

- Price: $165
- 64,000-word vocabulary, expandable to 250,000 words
- Adds voice navigation to most programs
- Easy to add new words

Minimum System Requirements
- 133 MHz
- 32MB of RAM
- 180MB of disk space
- Windows 95/98/NT 4.0

Dragon Naturally Speaking is among the most proven voice recognition software on the market, offering great results. However, it is the most sensitive to microphone placement.

Voice Xpress Advanced 4.0

- Price: $45
- 230,000-word vocabulary
- Recognition advisor helps maximize accuracy, directing the user during training
- Allows shortcuts, automatically inserting an entire section of text through a single command

Minimum System Requirements
- Pentium II processor
- 48MB of RAM
- 250MB of disk space
- Windows 95/98/NT 4.0

At under $50, Voice Xpress is among the best values in software. It is by far the best program to navigate all of the windows on your computer by voice, and if you only use it for that purpose, it's still a great product for the money.

Via Voice Pro

- Price: $169
- Specialized vocabularies, including business, computer and cuisine.
- Works in Web chat rooms.
- Functions with most Microsoft 2000 software versions (Word, Excel, Outlook and so on).

Minimum System Requirements
- 233-MHz Pentium processor
- 48MB of RAM
- 310MB of disk space
- Windows 95/98

For surfing the Web by voice, Via Voice is excellent, all but eliminating the need for a mouse and keystrokes.

Good Enough

While voice recognition software still has a few years to go before it reaches perfection, today's versions offer levels of accuracy that benefit many who have difficulty typing. It's not time to throw away your keyboard, but if you type fewer than 30 words per minute, you may want to pick up a headset.

(This story was posted on 2 Jun 2000)