The world’ s first speech-recognition system was the size of a shoebox
Virtual assistants have changed how we talk to computers. These AI-powered systems understand what we say and carry out tasks that secretaries used to handle: from dictation and scheduling to finding information.
The story begins with early speech recognition experiments. In 1952, Bell Labs built“ Audrey”, a machine the size of a six-foot relay rack that could recognise spoken digits with 90 % accuracy, though only for designated speakers.
Then William C. Dersch invented the IBM Shoebox in 1961 – the first speech-recognition system able to recognise 16 spoken words including digits 0-9 and basic commands, converting voice into electrical signals to operate an adding machine, pioneering early NLP.
The 1970s would bring“ Harpy” from Carnegie Mellon University, funded by DARPA, which understood 1000 words and could make sense of complete sentences. A major breakthrough came in 1997 when Dragon NaturallySpeaking
208 October 2025