Her voice was once described as sounding like two robot tobacco auctioneers fighting over a cigar butt. However, Audrey's synthesized speech was state-of- the-art back in 1952. A clumsy acronym for Automatic Digit Recognition, Audrey was an analog computer at Bell Laboratories. But it was not her primitive voice that she was renowned for. Scientists at Bell Laboratories had tried for many years to devise a technology that could recognize human speech and Audrey was their first working solution.
The potential applications for a computer that could convert words spoken by humans directly into digital text were obvious. However, the sheer number of variations in the qualities of people's voices, along with their different intonations and pronunciations, means that foolproof speech recognition is a huge task. Audrey could only recognize spoken numbers from one to ten, and used flashing lights to illustrate what she had heard—or thought she had heard. She compared the spoken sounds to the sound patterns stored in her memory and simply decided which were the most similar.
The technology was full of promise, and Bell Laboratories wanted to turn her into an operator that could dial telephone numbers for you. The problem was accuracy. Audrey had to be "tuned" for specific people's voices, and then she could only guarantee an accuracy of 98 percent, which is nowhere near good enough for commercial application.
Since the 1950s, the rapid increase in computing power has enabled many groups to experiment with their own speech recognition systems. One can even buy them for a home computer and program them with Specific phrases in order to play computer games without touching the keyboard.
