“Every building has its claim to fame,” says janet baker as she leads me around a three-story brick building that sits on a hill overlooking Boston. Once a mill, this building has been cleaned, renovated and turned into offices. Today it’s the headquarters of Dragon Systems, the company Janet and her husband Jim Baker founded in 1982.
“What’s this one’s?” I ask.
“The rope that hung John Wilkes Booth was made here,” she says with a smile.
Once I know the industrial building’s past, the signs are everywhere. The floors on the second and third floor are slightly tilted, so that workers a century ago could roll the massive spools of rope. There are doors on the third floor that open into empty space, where block and tackle lowered the spools to the carriages waiting below. Pulleys and rollers still hang from the building’s ceilings.
But historians looking back from the 21st century are less likely to remember this old millhouse for the noose that wrung the neck of Abraham Lincoln’s assassin than for being the place where Dragon Systems solved a “grand challenge” of computer science: getting a personal computer to recognize natural human speech.
Ever since the last century, engineers have been trying to build a machine that would heed its master’s voice; even Alexander Graham Bell tried his hand at it. And while computers capable of recognizing single spoken words have been around for decades, in the fall of 1995 pundits were still proclaiming that desktop machines capable of transcribing continuous speech-the rapid and sometimes muddled way people actually talk-wouldn’t be around until at least the year 2000…and possibly much later.
Today, you can buy Dragon Systems’ NaturallySpeaking at computer stores for $99.95 and run it on a new PC costing less than $2,000.
So just what can this technology do? Earlier this year I sat in a conference room at Dragon’s headquarters with a bunch of skeptical technology writers while Joel Gould, Dragon Systems’ lead architect, demonstrated the program he helped create. Gould walked to the front of the conference room, plugged his laptop into the projector, donned a lightweight telephone headset and started talking.
“I am going to give you a demonstration first, and then I will go back and show you some of the things that you saw go by quickly,” said Gould. A few seconds later the same words appeared on the screen, typed magically by the computer itself. Gould proceeded in this conversational style, with the machine transcribing everything he said. Although there was an occasional mistake, the machine’s accuracy was remarkable. Hoping to stump the program, a reporter asked if it could distinguish between words that sound the same but are spelled differently. Gould smiled, and let out a doozy: “Please write a letter right now to Mrs. Wright. Tell her that two is too many to buy.” The system recognized the words perfectly.
Dragon’s management confidently predicts that five years from now a computer without such voice recognition software is going to seem as primitive as a computer without a mouse would seem today. Letters and e-mail will be dictated as easily as talking on a phone. Just one step beyond that, PC-based simultaneous translation could topple language barriers.
Speech recognition’s arrival a few years ahead of schedule is largely due to the perseverance of Jim and Janet Baker, the couple who founded Dragon back in 1982. As researchers, the pair helped to invent some of the fundamental algorithms used today by all speech recognition products. As entrepreneurs, they fought to commercialize the technology years ahead of anyone’s schedule. Now that speech is on the desktop, it’s clear that our computing future will be shaped in no small part by Dragon Systems and the husband-and-wife team that gave birth to it.