Text of Seymour article (URL posted by advalorem):
Speech-Recognition Vendors Hear the Footsteps
By Jim Seymour
Special to TheStreet.com
7/6/00 7:01 AM ET
As I said in yesterday's column, what no one's talking about (in public, at least) is the real driver in the voice-recognition market right now: Fear of Microsoft (MSFT:Nasdaq - news - boards). (Remember in the months to come that you read it here on RealMoney.com first!)
For years, Microsoft has had a group of linguists, programmers and other smart people working on voice-recognition technology, and their work has produced really excellent results. I've worked with the Microsoft technology in the research labs in Redmond, and believe me, it's not only good, but a lot better than anything else I've tried. (No, unfortunately, the 'Softies wouldn't let me take a pre-prerelease copy home, so I still use Dragon's NaturallySpeaking. Sigh.)
Release of the Microsoft technology, in the form of a product you and I could buy, has been held back in Redmond for years over three big arguments:
1) Where does speech recognition really belong? In applications, or in the operating system?
2) How much computing power is enough to deliver true, real-time speech recognition?
3) Is this an antitrust tar pit for Microsoft, with serious repercussions in Washington, D.C.?
Where?
The first argument is as much philosophical as technical. If speech recognition were embedded in an operating system, such as Windows, PC users would have access to it no matter what they were doing at their computers -- in other words, not only for operating-system functions and commands, but within every program running under that operating system. (Or, at least, within every program compliant with the operating system vendor's API, or "application programming interface" spec for that product.)
Looking at directories of files, sending files and messages in an email program, dictating text into a word processor, entering tedious columns of numbers in a spreadsheet, searching in a database -- all would be "voice-enabled," for both content and commands.
If, on the other hand, speech recognition is viewed much more narrowly, and built only into a specific application or two -- say, Microsoft Word, in the Microsoft Outlook integrated-software bundle -- the PC user gets to use it only in that app (and with some system-level functions, such as file saves and printing, within that application).
I come down strongly on the side of putting it in Windows itself, because I can tell you, once you start getting used to dictating to your PC, you're going to want to be able to do that all the time. A solid speech interface has dozens of advantages ... and having them here but not there would be -- is -- exasperating.
Putting speech recognition into the operating system also means you have just one "training session," and custom vocabulary, which works across all apps, all commands. So the burden on the user is much less, too.
But those who argue for adding speech-recognition capability only to those products where its benefits are most obvious -- in other words, to word processing -- have a good point. They say we're far more likely to actually use the feature if we immediately see its benefits. And if it's not intimidating by virtue of a narrower scope.
(I understand. My 7-year-old Lexus has an extremely good voice-recognition system built in for the car's [analog] cell phone. Works really well, in good part thanks to the limited vocabulary it must deal with. But a Lexus person told me a couple of years ago that their research shows very few Lexus buyers ever try that feature, as convenient as it is, and as much as it helps with safety!)
PC Power?
It is fair to say that good speech recognition, in something vaguely like real time, requires a lot of computing horsepower. I struggled with Dragon's products until I got a 500-megahertz Pentium III-based PC, at which point everything fell into place. Microsoft wanted to wait until a sizable portion of the installed base of PCs -- or at least, many of the PCs used by the kind of customers likely to buy into speech recognition -- had enough power. Good decision.
Trustbusters?
Finally, Microsoft apparently now believes that it just couldn't get any worse on the antitrust front, so what the heck: Back to designing the best products it can, not worrying about How It Will Play in Washington ... and thus, this fall you'll see Microsoft-brand speech recognition in the newest version of Office.
The Microsoft Touch
Note that I'm not saying it'll be perfect -- remember the rule: Look for things to settle down in about the third release of any new Microsoft product -- but I expect to see very usable voice recognition in the new Office.
Well-informed investors with good memories will recall that Herb Greenberg's nemesis, Lernout & Hauspie (LHSP:Nasdaq - news - boards), licensed its speech-recognition technology to Microsoft more than three years ago, in November 1996. Does this mean what we'll see in Office this fall, four years after that deal, will be just a warmed-over version of L&H's technology?
No way. I assumed at the time that Microsoft did that licensing agreement to protect itself from patent-infringement claims, and also because there was some detail in the L&H interface it wanted to pick up. 'Softies won't confirm that, of course.
The advent of Microsoft-brand speech recognition would be enough to tip the apple carts of every other speech-product vendor. But add the close integration of speech recognition with the market-leading Microsoft apps in Office and... well, you get the picture. If 90%-plus of the more-or-less-serious users of PC "productivity" apps use Office -- and they do -- who's going to buy an external, inferior, add-in product for speech recognition, when the best comes free in the box...?
Whatever the story on Lernout & Hauspie's financials, whatever the quality of the technology they have bought, whatever IBM (IBM:NYSE - news - boards) wants to do with its good-but-not-that-good Via Voice technology -- I think the Microsoft move this fall is going to mark a permanent change in this market.
And I wouldn't want to be holding a competitor when that happens.
--------------------------------------------------------------------------------
A reader advises me that Lernout & Hauspie does have Korean-language versions of its speech-recognition packages, so that goes a little way toward explaining the puzzling report of the company's sales shooting up in Korea from $97,000 a year ago to $58.9 million in the first quarter of this fiscal year. Still...
The depth and breadth of knowledge of TheStreet.com and RealMoney.com readers never cease to amaze me... |
