Back to ComputerAmerica.com


Craig Crossman photo Craig Crossman
National Newspaper Computer Columnist

Click here to see the index of Craig's columns.

SPEECH TRANSCRIPTION FINALLY COMES TO THE MACINTOSH

The ability of a computer to accurately transcribe the spoken word into printed text has been a sort of Holy Grail to many software developers. To me, it never ceases to amaze me how the average person doesn't understand the complexity of such an undertaking. To most, having a computer understand what we are saying seems pretty straightforward and a natural application of man and machine. Perhaps this is due to all the movies and TV shows that depict characters speaking to computers and accepting their accurate understanding in a cavalier manner. But the reality is that the challenge is an enormous one and it continues to push the limits of artificial intelligence programming techniques as well as hardware power. In the past year or so, a quantum leap in the science of vocal transcription occurred when developers figured out how to get computers to recognize continuous speech. Up until then, you had to pause momentarily between each word.

Three major developers have emerged as leaders in this field. Dragon Systems offers Naturally Speaking, Lernout & Hauspie has Realspeak and IBM publishes ViaVoice. Competition between each of these companies is fierce and each new version of their respective programs produces added features and increased accuracy. The latter continues to be the most important aspect of speech transcription. After all, having the computer transcribe something other than what is being said is unacceptable. But it looks like IBM has scooped its competition by being the first to come out with its new ViaVoice Millennium Edition for the Macintosh.

Until now, the Mac has been mostly ignored by speech recognition companies. That's too bad because Apple was a short time leader in this arena with its PlainTalk technology built into the Macintosh operating system. This speaker independent recognition (It doesn't require training) didn't transcribe but it could be tailored to accept commands for just about any computer instruction such as opening a file or running a program. Unfortunately, lack of interest and industry support brought PlainTalk's further development to a screeching halt. But with ViaVoice for Mac, it looks like IBM was listening after all. Some might think it a bit ironical to have IBM as the company responsible for Apple's return to this market. But I don't care if it came from outer space. I'm just glad to see the Mac version's arrival on the planet.

ViaVoice for Mac comes with a quality headset with built-in microphone. It even comes with colored plastic snap on pieces to match any of Apple's popular iMac colors. The program walks you through a simple 20 minute session that trains ViaVoice for Mac to the sound of your voice and speaking cadence. Once done, you can immediately start transcribing to popular applications such as AppleWorks, Microsoft Word and Outlook Express, America Online's email and Netscape Messenger. Modeless operation lets you issue commands by simply pausing before the command or saying the word "computer." Editing and formatting is accomplished in the same manner by first pausing then saying things like "Capitalize this" or "Bold this." ViaVoice for Mac's natural dictation lets you speak numbers, dates, times measures and prices and have them properly notated automatically. A macro ability lets you type a complete phrase with one or two words such as "my address."

ViaVoice for Mac comes with a 260,000 word dictionary plus the ability to add your own words. Another nifty feature is ViaVoice for Mac's ability to add words to its dictionary from documents you have already typed. Bottom line is that the more you use ViaVoice for Mac, the more accurate it becomes.

ViaVoice for Mac requires an iMac, Power Mac, G3, G4 or PowerBook G3, Mac OS 8.5.1 or higher and is available now at retailers everywhere. $89.95.

IBM, Corp.
www-4.ibm.com/software/speech/mac/

Craig Crossman is a national newspaper columnist writing about computers and technology. He also hosts the number one computer radio talk show, Computer America, heard on the BusinessTalkRadio.Net, every Sunday, 3-6PM, ET. In South Florida, "The Craig Crossman Show" is heard Sunday evenings from 10 to Midnight on WJNO-AM1040, WJNX-AM1330 and WBZT-AM1290.

Back to ComputerAmerica.com