Microsoft Invests in Voice Recognition

September 1997

Microsoft made an investment in the voice recognition market today, buying a 16% interest in Lernout & Hauspie for $45 million. This is in the nature of a strategic alliance between the two companies, designed to speed up Microsoft’s entrance into the speech market and L&H’s presence and importance in that market.

In the short term, this alliance is mainly about providing co-marketing for L&H’s dictation products, acquired from Kurzweil in June. These include both discrete word dictation and command-and-control products currently in the market and a continuous speech product expected this fall.

Exactly how this joint marketing will occur is still uncertain, L&H CEO Gaston Bastiaens told us, since the details of the deal are still being finalized, but products marketed under Microsoft’s oh-so-broad marketing umbrella are sure to attract more market attention and sales than Kurzweil/L&H alone. That’s important, because the other major player in the field is IBM, who has brought the sleepy voice recognition business much attention by some sharp pricing actions, starting last fall.

Now, with much more usable continuous speech dictation products from Dragon Systems and IBM in the market, things have really heated up. We’d guess this made Microsoft anxious to be a player in the game now (and L&H happy to return to its role as a provider of core technologies and oem solutions).

In the longer run, L&H and Microsoft are forming a joint venture in Flanders (Belgium, where L&H has its European headquarters) to gather voice data; millions of document and voice samples are required to build accurate and reliable speech engines. Microsoft will also invest $3 million in a Belgium-based language research consortium and in a Belgian computational linguistics research program.

Microsoft is expected to use L&H’s continuous voice recognition engine, together with its own voice research, to embed voice interface technology into the Windows operating system. Pressed for a time schedule, Bastiaens demurred, but we’d guess you shouldn’t expect this much before the next release of Windows/NT – not Windows 98, but rather Windows 2000. This makes sense. We doubt the problem is how to implement voice processing – that’s almost here – but rather the issue of how many people could run it:

• Today, continuous speech voice recognition requires a 166 MHz Pentium processor and 32 MB of memory, in a world where there are lots of older, slower computers.
• Many people still use Windows 3.1 rather than Windows 95 because their computer is literally too small; imagine how many fewer could run a Windows OS which included voice.
• No problem, you say. It would just be an option. But it’s time to rethink the interface and an interface optimized for voice is inherently different than one optimized for a mouse or other pointing device. You don’t want a mismatch between a newly upgraded operating system and its new interface!
• By the time Windows 2000 ships (beta sometime in 1999, we’d guess), this problem will be much smaller, as many more old machines will have been replaced by larger machines with more memory.

Of course, some of these machines may be replaced by Network Computers (NC’s). These thin clients probably couldn’t run a fat application like voice at all, but there’s nothing to prevent the recognizer from being a multi-user server-side application, with a client/interface on the NC. We not only think this would work, we expect to see this first (before major changes in voice enabling the OS itself).

This leads to our final conclusion. Microsoft is interested in voice in a very broad way – just as it is interested in the Internet and on-line services and content in a very broad way. Note that their first relationship with L&H was for voice technology for its SAPI/telephony business. And that’s where we expect to see a continuous speech, voice-enabled generic interface pop up first. The telephone is a natural server-centric architecture and voice is its normal interface. Telephones are, by definition, thin clients. The new smart phones coming on both the Java (announced at JIBE) and Windows/CE (to be announced any minute) platforms will be chubbier, but they’re unlikely to be able to process continuous speech. We’d not be at all surprised to see Microsoft and its new partner L&H pioneering here.

Some are very disappointed to see Microsoft looking at continuous speech and voice interfaces as part of the Windows/NT environment. Especially the voice interface vendors who have newly brought this technology to market like Dragon Systems and IBM. But:

• They have a substantial window (we’d guess at least several years) in which such function will only be available as a separate application, in which to build a large, if temporary market.
• They have the opportunity to build significant businesses in the incremental voice products that will sit on top of a generic continuous speech processing function – as they always intended to do – and as L&H plans. It’s a smaller market, but a lucrative one.
• We suspect that – as always – the prizes here will be awarded not for best technology, but rather for best marketing and most flexibility in adapting to a rapidly changing market environment in which Microsoft now intends to play a significant role. Much of the future business (post-voice in the OS) will be in selling incremental voice function to oem’s for commercial and custom applications so building relationships now is critical.

As you know, we are working on a White Paper on Voice Processing, with a tentative publication date of late October. This and other developments will, of course, be incorporated into our business models. Perhaps what we need is not a White Paper, but a continuing dialog, and we’re thinking about that, too. For information on our White Paper please visit our web site at http://www.wohl.com.

Comments or Questions: Send Email to opinions@wohl.com

Home/ Search / 2005 Articles / Issue Archive / Free Newsletter

Entire contents © 1997 by Amy D. Wohl. All rights reserved. Reproduction of this publication in any form without prior written permission is forbidden.