When did machines first start recognizing human speech - and how many words could they understand?

The answer

Automatic speech recognition began in 1952 at Bell Labs, where researchers Stephen Davis, Karl Davis, and R. Biddulph built 'Audrey,' a system that could recognize spoken digits (zero through nine) from a single speaker with about 97% accuracy. It worked by matching the spectral properties of incoming speech to stored reference patterns. Progress was slow: by 1969, John Pierce of Bell Labs wrote a famous letter to the Journal of the Acoustical Society of America discouraging further speech recognition research. The breakthrough came in the 1970s when James Baker at Carnegie Mellon and Frederick Jelinek at IBM independently applied hidden Markov models (HMMs) to speech, shifting the field from hand-crafted rules to statistical methods. DARPA's Speech Understanding Research program (1971–1976) produced Carnegie Mellon's Harpy system, which could understand 1,011 words from connected speech.