The phonetic typewriter
In November 1956 Harry F. Olson and Herbert Belar of RCA Laboratories described a phonetic typewriter, a simplified but nevertheless complete system that types in response to words spoken into a microphone. The paper works through five problems: the form in which the words are typed, the means of analysing the sounds of speech, the identification of the analysed sounds, the encoding and decoding that drives the actuating mechanism, and the design of that mechanism.
Why it matters
For the first time the output of a listening machine was text on paper rather than one signal picked from a short list. The same work showed what such a machine breaks on: the spectrum was written into memory in fixed increments of time, so the moment a talker changed pace the record slid out of register.
The full text of the 1956 paper is paywalled and its abstract carries no figures. That the machine did type in response to spoken words, and that it transferred the spectrum into memory in fixed time increments, comes from a separate June 1959 publication by the same authors which the publisher serves free and which describes both machines. The same item describes the 1959 extension itself: coincidence detectors compare the current speech spectrum with the preceding one, and nothing is fed into the spectral memory until a significant change has been detected. What the record does not claim: how many band channels the analyser had, or the size of the vocabulary. The research report behind this batch gave eight channels and ten syllables; neither appears in any accessible abstract.