Dragon Systems

Dragon Systems was an American speech recognition company based in Newton, Massachusetts ^[1]. It built software that turned spoken words into text on personal computers, and its products — DragonDictate first, then NaturallySpeaking — are part of how dictation moved from laboratory demonstration to something a person could buy.

Where it came from

James K. Baker had already laid out a description of a speech understanding system called DRAGON in 1975 ^[1]. In 1982 he and his wife Janet M. Baker founded Dragon Systems in order to release products based on their voice recognition prototype ^[1]. The division of roles stayed fixed: James Baker served as president of the company and Janet Baker as chief executive ^[1].

The products

The first product was DragonDictate, released for MS-DOS and built on hidden Markov models ^[1]. Technically it rested on a trigram model and is known as a discrete utterance speech recognition engine ^[1]. The limiting factor was hardware: the machines of the day could not yet handle word segmentation, so a user had to enunciate one word at a time, leaving a small pause after each word ^[1]. BBC Chinese reports that DragonDictate was launched in 1990 as the first consumer speech recognition product, at a price of USD 9,000 ^[2].

The step from single words to running speech came with NaturallySpeaking 1.0, released in 1997 as the company's first continuous dictation product ^[1]. That contrast — discrete utterance versus continuous dictation — is the clearest way to see what changed over the company's life: the earlier system asked the speaker to adapt to the machine, one word and one pause at a time, while the later one was meant to follow speech as it is normally produced.

Change of ownership

In June 2000 the company was purchased by Belgium-based Lernout & Hauspie ^[1]. That November Lernout & Hauspie declared bankruptcy, after being found to have perpetrated financial fraud ^[1]. After the bankruptcy the rights to the Dragon product line were acquired by ScanSoft, which in 2005 acquired Nuance Communications and rebranded itself as Nuance ^[1]. The founders did not let the sale pass quietly: the Bakers sued Goldman Sachs over the all-share deal, and a 23-day trial in Boston in January 2013 ended with the jury clearing Goldman Sachs of all charges ^[1].

Where it sits

The history of Dragon Systems is a compact example of a common pattern in applied speech research. A technical idea is described publicly years before there is any product — the DRAGON description of 1975 ^[1] — and the company is then formed to turn a prototype into something sellable ^[1]. Progress shows up as a sequence of concrete engineering concessions: hidden Markov models and a trigram model to do the recognition ^[1], a rigid one-word-at-a-time speaking style forced by hardware that could not segment words ^[1], and eventually a continuous dictation product ^[1].

The commercial side is just as instructive. The company itself disappeared in the Lernout & Hauspie bankruptcy and the fraud that preceded it ^[1], but the product line survived the collapse of its owner, passing through ScanSoft and ending up inside Nuance ^[1]. The name on the box changed; the technology kept going. And the price attached to the first consumer product, USD 9,000 ^[2], is a reminder of how narrow the market for speech recognition was at that point, long before dictation became a feature people expect to find built into ordinary software.