Two Bell Labs sources disagree on the first singing computer

The system

In the early 1960s, researchers at Bell Telephone Laboratories made a computer sing 'Daisy Bell' ('A Bicycle Built for Two'), a demonstration usually cited as the first computer-synthesized singing voice. The Library of Congress National Recording Registry credits the recording, dated 1961, to Max Mathews, John L. Kelly Jr. and Carol Lochbaum, and added it to the registry in 2009. A separate Bell Labs background page on its own text-to-speech history frames the same demonstration as Kelly's work, with Mathews providing musical accompaniment.

What the documents establish

The two institutional sources do not agree on every detail. The registry entry gives the year as 1961 and names all three researchers as principal talent, without specifying the computer used. The Bell Labs page instead dates 'the sample created by John L. Kelly' to 1962 and identifies the machine as an IBM 704, run through Kelly's vocoder synthesizer. Neither page is a contemporaneous lab notebook; both are institutional retrospectives written years after the fact, which is itself worth noting whenever an anniversary claim states a computer model or exact year with more confidence than the sources actually share. What both agree on is the outcome: a synthesized voice performing the song with computer-generated musical accompaniment, and a later connection to popular culture, since the Bell Labs page records that Arthur C. Clarke saw the demonstration while visiting colleague John Pierce at the Murray Hill facility and later used it for HAL 9000's death scene in 2001: A Space Odyssey.

Craft and rights

This is a research demonstration, not a released product, so no licence or dataset question attaches to the recording itself. The more durable craft point is that voice synthesis and speech synthesis shared laboratory equipment and researchers from the start; Bell Labs' own history traces a line from the 1939 Voder speech synthesizer through this 1960s singing demonstration to later text-to-speech systems. Any modern claim about a voice model's originality should reckon with a lineage this old and this well documented, and with how much of that lineage exists only in institutional summaries rather than raw technical records.

Outcomes and open questions

Even the Library of Congress page is internally inconsistent, spelling the third researcher's surname as both Lochbaum and Lockbaum in different tables on the same listing. Combined with the year and computer-model disagreement between sources, the episode is a caution against citing a single tidy fact for an event this old without checking whether the institutions that hold the record agree with each other.

• Do two institutional accounts of the same historical demonstration actually agree on the year, the hardware and the people involved?
• What changes about a provenance claim when the only surviving sources are retrospective institutional pages rather than contemporaneous documentation?
• How much of a synthetic-voice origin story is repeated because it is checkable, versus because it is dramatic?

The safest statement the record supports is a narrow one: Bell Labs researchers produced a computer-synthesized rendition of 'Daisy Bell' around 1961, credited to Mathews, Kelly and Lochbaum, and the exact year and machine still depend on which institution's account is being read.