The IBM Shoebox was an experimental voice-recognition device built at IBM in 1961 and demonstrated to the public at the 1962 World’s Fair in Seattle. Roughly the size and shape of a shoebox, it was developed by IBM engineer William C. Dersch in the company’s Advanced Systems Development Division laboratory in San Jose.
According to IBM’s own archives, the machine “would recognize 16 words spoken into its microphone and convert those sounds into electrical impulses.” The 16 words were the digits zero through nine plus six command words, including “plus,” “minus,” and “total.” Speaking a sum out loud - for example, the digits of two numbers followed by “plus” and “total” - caused the Shoebox to drive an attached adding machine and produce the answer. A row of small lamps labeled zero to nine lit up to show which digit it had heard.
Like Bell Labs’ Audrey a decade earlier, the Shoebox worked on a small, fixed vocabulary and depended on clear, deliberate speech. It used analog filters to classify sounds rather than any stored program. It was a demonstration of possibility rather than a shipping product, but it put speech recognition in front of a general public audience for the first time and became one of IBM’s signature early-AI showpieces.
For business readers, the Shoebox is an early example of a company using a striking technology demo to shape its public image. The hard problems - large vocabularies, multiple speakers, continuous natural speech - would take another fifty years and a shift to statistical and then neural methods to solve at consumer scale.