Hello, computer?

I’ve been testing voice recognition software for at least five years. I haven’t written about any packages because I haven’t found one that works well enough for me to use for more than a few minutes without becoming frustrated.

That day may be coming, however, with the release of the latest generation of speech recognition products. Lernout & Hauspie’s (L&H) VoiceXpress 4 Professional, which came out this month, “typed” 120 to 130 words per minute with an average of nine recognition errors when I tested it on a 450-MHz Pentium III with 128M bytes of memory. Compare that to my typing about 60 words per minute, with three errors.

I was fairly happy with the speed and accuracy, especially considering you can improve the accuracy several ways. First, you have to reduce the product’s sensitivity to background noise. If you don’t, it interprets random sounds as random words when you’re not talking and mangles other words when you are. Keep the microphone away from white noise sources such as fans. Each time you sit down to dictate, tune the microphone.

Second, make sure you train the software to recognize your voice. L&H calls this process “enrollment.” The company provides several passages for you to read into the microphone. Each takes about 10 minutes. Once you’re done, the computer processes your speech to create internal models for your pronunciation. That process took about 75 seconds on my computer.

I suggest enrolling more than once; I found that recognition improved after I read a second passage.

You can improve accuracy further if you train the program as you go along. If VoiceXpress enters an incorrect word, you can highlight it and choose from a list of alternate words that you might have meant, or type in the correct one. If you click on the Train button, the program makes you pronounce the new word twice, after which it does a better job understanding it in the future. You can use a similar process in advance for any unusual words you use frequently. I noticed improvement the longer I worked with and trained the software.

I found VoiceXpress works better for dictation tasks than for editing documents. My tendinitis would have to turn into full-blown carpal tunnel syndrome before I’d edit using voice commands alone.

I don’t have much room to touch on the product’s cool features for executing complex commands by making simple statements such as “number the next two paragraphs,” nor space to detail how it works with other Office applications, not just word processing. Though VoiceXpress 4 Professional can enter words faster than I can type, it can’t find more space for my column.

No Comments

No comments yet.

RSS feed for comments on this post. TrackBack URI

Leave a comment

WordPress Themes