In the autumn of 1951, a BBC outside broadcast unit entered the computing laboratory at the University of Manchester with a portable disc cutter and recorded something that had never been captured before: a computer playing music. The machine was the Ferranti Mark 1, which Alan Turing referred to as the Mark II, a cabinet of valves that filled a room. It played three pieces: the British national anthem, Baa Baa Black Sheep, and the Glenn Miller song, In the Mood. The latter two were unsuccessful. The second remained a half-finished attempt, and during the third, the computer crashed. On the acetate, one can then hear the voice of the presenter, who resolves the failure with a perfect phrase: the machine, quite obviously, is not in the mood.
That disc is the oldest known recording of music produced by a computer. Not the first computer music, which was produced earlier in that same laboratory and also in Australia, but the oldest that has been preserved as a recording. And for decades it was heard as it had been preserved, with its tones, its interruptions, the announcer's voice, and the metallic hum of the notes. Until two researchers, Jack Copeland and Jason Long, set out to analyse it and discovered that what we were hearing was not what had sounded in the room.
The discovery possesses an elegance that warrants explanation. The Ferranti Mark 1 could not produce just any note. Its loudspeaker emitted a click every time the machine executed a specific instruction, and by repeating those clicks at a constant rhythm, a tone was obtained. But the rhythm depended on the computer's clock, so it could only generate a handful of specific frequencies, those resulting from its cycles. It could not tune. When it attempted to play a G at 196 hertz, the closest it could reach was 198.41. The machine played out of tune by design, and that detuning was calculable: one simply had to read Turing's manual and perform the calculations.
Copeland and Long performed those calculations and then measured the frequencies on the disc. And there the problem appeared: many of the recorded notes did not coincide with any of those the computer could produce. It is not that they were out of tune with respect to the musical scale; it is that they were impossible, frequencies that that machine was incapable of generating in any way. Something, between the laboratory and the disc, had shifted the entire sound.
The most probable explanation, and the one they present with caution, is that the turntable of the portable cutter rotated too quickly while the acetate was being recorded. By cutting the groove at an excessive speed, the waves were stretched along the disc. Later, when played back at the standard speed of seventy-eight revolutions, the needle traversed those waves more slowly than they had been recorded, and everything sounded lower and deeper than it had in reality. A minor mechanical failure, a motor running slightly fast for a few minutes in 1951, and sixty-five years of people listening to notes that the machine never actually played.
What is remarkable is how they were able to correct it. They did not do it by ear, nor by tuning until it sounded correct. They did it in reverse: since they knew exactly which frequencies the computer could produce and which it could not, they sought the speed factor that returned all the recorded notes to the set of possible ones. When they accelerated the playback by that exact proportion, the impossible frequencies disappeared and all of them fit into those the machine could indeed produce. Afterwards, they filtered out the extraneous noise and corrected with software a speed oscillation that the disc cutting itself had introduced. The technical limitation of the computer, its inability to tune, was what allowed it to be reconstructed: because it could only play a few notes, the notes it could not play revealed the error.
The researchers speak of the true sound of the machine, and their enthusiasm is understandable, but what exists is a technically grounded restoration, not proof of acoustic identity. There is no other reference recording, nor a calibration signal recorded that afternoon, nor a measurement of the actual speed of the cutter. What we have is the best possible reconstruction of how it must have sounded, supported by a solid calculation rather than an intuition. It is a great deal, and it is honest to state that it is that and nothing more.
I am drawn to the paradox of the medium. That acetate disc performed its function: it preserved the performance, the sequence of notes, the interruptions, the computer failure, the voice of the woman joking. None of that was lost. What it did not preserve accurately was the pitch of the sounds, precisely what one would assume to be the easiest element to retain in a recording. For decades, we were presented with a document that was faithful in almost every respect yet displaced in the essential, and no one noticed, for there was nothing with which to compare it. Only when someone calculated what that machine could and could not do did the recording itself confess its error. Media neither lie entirely nor tell the whole truth: they preserve certain elements while distorting others, and determining which requires knowledge that the medium itself does not contain.
On the open conversation
This text takes as its starting point a verified case—the 1951 BBC recording of the Manchester computer playing music and its subsequent restoration by Jack Copeland and Jason Long—to consider what a recording preserves and what it distorts. I distinguish between the facts (the disc, the impossible frequencies, the method of correction) and what is probable (the cause of the displacement, which the researchers themselves present with caution) and what would be excessive (to claim that we hear today exactly what sounded in the room). I also clarify that the recording was not discovered in 2016; it was already preserved: what occurred then was its restoration. This is the second text in this notebook concerning sound. Should anyone wish to contribute from the fields of electronic music, the history of computing, or sound preservation, this notebook remains open.
Sources
B. Jack Copeland and Jason Long, 'Turing and the History of Computer Music' (Springer, 2017; corrected open-access version, 2021), authors of the analysis and restoration.
British Library, 'Restoring the first recording of computer music' (originally published in September 2018), featuring the restored audio; digital preservation copy used, ref. H3942.
University of Manchester, 'First digital music made in Manchester', regarding the session and the provenance of the acetate (a copy cut for the engineer Frank Cooper, later held by the Computer Conservation Society and the National Sound Archive).
The restoration was released to the public in September 2016. The recording was already known: BBC News published a report on it on 17 June 2008.
Clarifications: the machine was the Ferranti Mark 1 (Turing called it the Manchester Electronic Computer Mark II); Turing explained in his manual how to produce notes using the loudspeaker instruction, but he did not program the melodies; Christopher Strachey programmed the first documented performance of the anthem, and the authorship of the three recorded routines remains unresolved.