Speech recognition has come a long way in 50 years

Excerpt from the March 4, 1967, issue of Science News

talking to a smartphone

SAY WHAT?  Hello, Siri, can you understand me now? That conversation goes much differently today than it would have in the 1960s, when devices only knew nine words.


Computers that hear

Computer engineers have dreamed of a machine that would translate speech into something that a vacuum tube or transistor could understand. Now at last, some promising hardware is being developed…. It is still a long way from the kind of science fiction computer that can understand sentences or long speeches. — Science News, March 4, 1967


That 1967 device knew the words one through nine. Earlier speech recognition devices sliced a word into segments and analyzed them for absolute loudness. But this machine, developed by Genung L. Clapper at IBM, identified the volume of a pitch segment compared with its neighbors to account for the variability of human speech. Today’s speech recognition goes much further, dividing words into distinct units of sound and syntax. The software decodes speech by applying pattern recognition and a statistical method called the hidden Markov model to the sounds. We rely on speech recognition to open an app to order groceries or to send a text to ask someone at home if we need more milk. Hello, Siri.

Bethany was previously the staff writer at Science News for Students. She has a Ph.D. in physiology and pharmacology from Wake Forest University School of Medicine.

More Stories from Science News on Computing