Latest News

L&H going down

So much for the leader in speech recognition software.

A few years ago, I attended a Lernout & Hauspie demonstration at Comdex. I was fairly green at the time. I had yet to taste the bitterness of deception that I have since found all too common in trade show hype. All that was about to change.

The demonstration was supposed to show that L&H’s new speech recognition software not only recognized human voice commands, but understood them. Led by co-founder Pol Hauspie, the demonstration was obviously faked. It looked breathtaking at first. But when the software script that was supposed to simulate real voice recognition got out of whack with the demonstrator’s commands, the gig was up. My fellow journalists and I left the demo room in disgust.

Last year, as the news wire streamed one story after another of the company’s apparent fraudulent behavior, I was reminded of the faked demonstration. I never really got over the trashing my naïvete took that day. Why would the envy of every other software company in that niche go to such lengths to try to fool a room full of journalists? And why would the only remaining speech-software company of any size (after it acquired Dragon Systems) thumb its nose at the financial world and fake its financials to such great lengths?

Well, news that ZD Net pulled down from Reuters last week has served to strengthen my opinion that the company was led by a couple of crooks. Both of the founders–Hauspie and Jo Lernout–have been thrown in a Belgian jail pending hearings to determine the extent of their criminal behavior. And it seems they won’t be getting out any time soon.

The bankrupt company is awash in debt–so much so that its considerable assets (we’re talking about 70 percent market share in speech products) won’t cover its liabilities. It’s being sued by investors of two Nasdaq markets. And several other pending legal inquiries don’t look good for the once vaunted L&H. The evidence that the cofounders kept dual books is overwhelming; and some have suggested that what was uncovered last year is just one instance in a long pattern of deception. It will take years to sort it all out in court.

Not since Michael Milken’s junk bond scam have we had such a high-profile fraud case. But I still don’t understand why two pioneers in this vital software sector would go so far astray. In the case of the faked demo, my colleagues and I did not expect the software to understand human speech. We would have been happy with a fairly simple demonstration that showed the software’s more modest, but real, capabilities. Apparently Hauspie was not satisfied. He wanted his software to do the impossible.

We users also would have been happy plugging along towards 99 percent accuracy by 2005, and buying the software as it approximated our needs. But Hauspie and Lernout were not satisfied with such meager success. So they cheated, and as Nelson King showed in an earlier column, they are not the only ones who will pay the price.

James Mathewson is editorial director of ComputerUser and

Leave a comment

seks shop - izolasyon
basic theory test book basic theory test