Connect with Fast Company

The international soldiers couldn't afford to continue their painstaking, frustrating interview process one day longer. Every hour the weapons specialists and language experts spent sweating out crucial conversations with Bosnian war survivors, more innocent lives were put at risk by the yet-undetected, live land mines scattered dangerously close to civilian camps.

Their task: Question the men, women, and children injured by enemy land mines in order to detect and disarm the weapons as quickly as possible. Their problem: The necessary two-member interview teams were expensive, slow, and often unsuccessful. Their solution: An innovative speech recognition company from Newton, Massachusetts, called Dragon Systems, Inc.

An unlikely military partner, Dragon was founded by a dedicated statistician who had unlocked the door to computerized speech recognition a decade earlier. Dragon didn't specialize in direct speech translation, and it certainly never before tailored a product for a war-ravaged society on the other side of the globe. Regardless, the assignment intrigued Dragon and its founders, James and Janet Baker. They accepted the challenge.

Within a matter of months, Dragon had developed a hand-held translation device equipped with a microphone, small display screen and speech recognition technology that instantly transformed English questions into Bosnian text. Armed with 10,000 possible questions and their translations, the tool allowed the Bosnian subjects to respond to soldiers with simple "yes" and "no" answers. Military manpower and timetables were significantly reduced. Lives were saved. And Dragon was recruited to create a similar device for use in Kosovo refugee camps.

"The Bakers are not your average blue suit and stripes business people -- they are techies with good hearts," says Dragon System's Director of International Sales and Marketing Andreas Widmer. "We didn't really make any money with this, but this is the right way to use the technology."

Since its inception, Dragon has colored outside the lines. When founder James Baker first proposed his thesis for using mathematical models and processes to enable machine speech recognition, he was shunned by engineers and thinkers at IBM, Bell Labs, and elsewhere. They claimed artificial intelligence was necessary. Baker said Markov's Model could be used to predict speech from statistical standpoint. And then he proved it.

One year after founding Dragon Systems, Inc., Baker unveiled a product that recognized 1,000 spoken words -- ten times the number distinguished by the cynics' systems. The following year, Dragon products could recognize 10,000 words. Needless to say, Baker felt vindicated.

During the subsequent decade, Dragon continued to introduce groundbreaking technology, all the while sticking close to four guiding principles and priorities: innovation, accuracy, usability, and efficiency -- "Dragon is streamlining algorithms to fit on smaller and smaller chips so that one day you will be able to speak to your watch," Widmer says.

"In 1997 we claimed the holy grail in speech recognition, and that was to have large vocabulary continuous speech products where the machine understands everything I'm saying and writes it down in real time," he says. "The original predictions were that a product like this could not be invented until 2005."