When did computers first type what you say?

A Bell Labs system recognized spoken digits in 1952 ¹, but only from one voice, and it did not type words. DragonDictate, launched in 1990, was described by its maker as the first commercially viable speech typewriter ². It needed a short pause between words ³. Software that handles ordinary, continuous speech arrived with Dragon NaturallySpeaking in 1997 ⁴. Phones followed in the late 2010s. The answer depends on what you count as "typing," so the dates below separate each step.

Our guide to the history of dictation covers the whole path, from scribes to phones. This page answers only the question of when a computer first typed what you said.

Computers recognized spoken digits in 1952, not words

The 1952 work was built for digits. In 1952, Davis, Biddulph and Balashek of Bell Laboratories built a system that recognized isolated spoken digits from a single speaker ¹. The Computer History Museum's account of AUDREY, the Bell Labs machine, says it could recognize the digits zero to nine with 90 percent accuracy, but only when its inventor spoke ⁵.

Two limits define this early work. The machine handled spoken digits from one person, with a vocabulary of ten digits. It was a research demonstration, not a writing tool. Recognizing a digit is a long way from typing a sentence.

IBM's voice typewriter was trained for each user

In the 1970s, IBM worked toward a machine that would type what a person said. Juang and Rabiner's history of the field describes IBM's effort, led by Fred Jelinek, as an attempt at a "voice-activated typewriter" ⁶. Their system, Tangora, was a research prototype rather than a product, and the authors say that "the typewriter had to be trained by each individual user" ⁷. A person had to train the system before it worked for them.

This matters for the timeline. A typewriter that needs training for each writer is a different product from one that works for anyone who walks up to it. The research aimed at a voice typewriter, and the systems that reached buyers came later.

DragonDictate in 1990 was described as the first commercially viable speech typewriter

A March 1990 newspaper report said that Dragon Systems of Newton, Massachusetts, had introduced what it said was "the first commercially viable typewriter that operates by human speech" ². The phrase "what it says is" is the company's claim, not a result of independent testing. Still, the report describes a product that had been introduced, so 1990 is the earliest date this page can support for a commercial speech typewriter.

DragonDictate had a hard limit. It recognized "discrete utterances," which the report defines as words with at least a quarter-second pause between them ³. A speaker had to say one word, pause, and then say the next. Continuous speech came later, as the next section shows.

Continuous speech reached writers in 1997

The jump from separated words to normal speech came with Dragon NaturallySpeaking. In April 1997, Dragon Systems unveiled it as the first general-purpose, large-vocabulary continuous speech recognition product ⁴. Removing the pauses between words was the main change. Sources differ on the retail release month for this product, so the year is firmer than the month.

Phones moved recognition onto the device in the late 2010s

The last step was where the work happened. In 2019, Google announced "an end-to-end, all-neural, on-device speech recognizer to power speech input in Gboard" ⁸. The post says the first rollout was for American English on Pixel phones. Apple describes a similar arrangement for iPhone: Dictation there is processed on the device in many languages, and no internet connection is required ⁹.

"On device" is not the same everywhere. Whether dictation works offline depends on the language, the operating system version and the settings you chose. Check the speech settings on your own device. Our answer on whether dictation works without internet goes through the cases.

What the dates mean if you want to dictate today

If you ask when computers first typed what you say, the answer is 1990 ² for the first speech typewriter sold, as its maker described it, and 1997 ⁴ for software that handles normal sentences. The 1952 system ¹ matters as an early recognizer, but no reader today can use it.

For practical use, three points follow from the history:

1. Speak in sentences, not separated words. The 1990 product required pauses ³, and the 1997 product was the first to avoid them ⁴. Modern tools expect the normal pace of speech.
2. Expect to train or correct. The 1952 system worked for one voice ⁵, and the research system Tangora needed training for each user. Current tools adapt more, but accuracy still varies by voice and subject. The question on whether you need to train dictation to your voice covers this in detail.
3. Use a tool built for your platform. Voice typing on Mac and voice typing on Windows are free, built-in starting points. Yap, which we make, is an on-device dictation app for Mac and Windows, and you can see our downloads if you want a dedicated option.

Speech to text and dictation are related but not identical. The question on whether speech to text is the same as dictation explains the difference, and the voice typing entry defines the term. For the older machines that recorded speech for a typist, see what a Dictaphone is.