13 September 2016
Restoring the first recording of computer music
Jack Copeland FRS NZ and Jason Long write:
A key problem facing audio archivists is how to establish the correct pitch of a historical recording. Without some independent means of knowing how the original sounded, it can be very difficult—or even impossible—to tell whether an archived recording is playing at the right pitch. An important case in point is the earliest known recording of computer-generated music. In 1951, a BBC outside broadcast unit in Manchester used a portable acetate disc cutter to capture three melodies played by a primeval computer. This gigantic computer filled much of the ground floor of Alan Turing's Computing Machine Laboratory.
Today, all that remains of the recording session is a 12-inch single-sided acetate disc, cut by the BBC's technician while the computer played. The computer itself was scrapped long ago, so the archived recording is our only window on that historic soundscape. What a disappointment it was, therefore, to discover that the pitches were not accurate: the recording gave at best only a rough impression of how the computer sounded. But with some electronic detective work it proved possible to restore the recording—with the result that the true sound of this ancestral computer can be heard once again, for the first time in more than half a century.
Alan Turing's pioneering work, in the late 1940s, on transforming the computer into a musical instrument has largely been overlooked: it's an urban myth of the music world that the first computer-generated musical notes were heard in 1957, at Bell Labs in America.1 The recent Oxford Handbook of Computer Music staked out a counterclaim, saying that the first computer to play notes was located in Sydney, Australia.2 However, the Sydney computer was not operational until the end of 1950, whereas computer-generated notes were emerging from a loudspeaker in Turing's computing lab as early as the autumn of 1948.
The Manchester computer had a special instruction that caused the loudspeaker—Turing called it the 'hooter'—to emit a short pulse of sound, lasting a tiny fraction of a second. Turing said this sounded like 'something between a tap, a click, and a thump'. Executing the instruction over and over again resulted in this 'click' being produced repeatedly, on every fourth tick of the computer's internal clock: tick tick tick click, tick tick tick click. Repeating the instruction enough times like this caused the human ear to hear not discrete clicks but a steady note, in fact the note C6, two octaves above middle C.
Turing realized that if the 'hoot' instruction were repeated not simply over and over again, but in different patterns, then the ear would hear different musical notes: for example, the repeated pattern tick tick tick click, tick tick tick tick, tick tick tick click, tick tick tick tick produced the note of C5 (an octave above middle C), while repeating the different pattern tick tick tick click, tick tick tick click, tick tick tick tick, tick tick tick click, tick tick tick click, tick tick tick tick produced the note of F4, four notes above above middle C—and so on. It was a wonderful discovery.
Turing was not very interested in programming the computer to play conventional pieces of music: he used the different notes to indicate what was going on in the computer—one note for 'job finished', others for 'digits overflowing in memory', 'error when transferring data from the magnetic drum', and so on. Running one of Turing's programs must have been a noisy business, with different musical notes and rhythms of clicks enabling the user to 'listen in' (as he put it) to what the computer was doing. He left it to someone else, though, to program the first complete piece of music.
A young schoolteacher named Christopher Strachey got hold of a copy of Turing's Programmers' Handbook for Manchester Electronic Computer Mark II (the Mark II computer had replaced the prototype Mark I, which also played notes, early in 1951).3 This was in fact the world’s first computer programming manual. Strachey, a talented pianist, studied the Handbook and appreciated the potential of Turing's terse directions on how to program musical notes. Soon to become one of Britain's top computer scientists, Strachey turned up at Turing's Manchester lab with what was at the time the longest computer program ever to be attempted. Turing knew the precocious Strachey well enough to let him use the computer for a night. 'Turing came in and gave me a typical high-speed, high-pitched description of how to use the machine', Strachey recounted; and then Turing departed, leaving him alone at the computer's console until the following morning.4
'I sat in front of this enormous machine', Strachey said, 'with four or five rows of twenty switches and things, in a room that felt like the control room of a battle-ship.'5 It was the first of a lifetime of all-night programming sessions. In the morning, to onlookers' astonishment the computer raucously hooted out the National Anthem. Turing, his usual monosyllabic self, said enthusiastically 'Good show'. Strachey could hardly have thought of a better way to get attention: a few weeks later he received a letter offering him a job at the computing lab.6
The BBC recording, made some time later the same year, included not only the National Anthem but also an endearing, if rather brash, rendition of the nursery rhyme Baa Baa Black Sheep as well as a reedy and wooden performance of Glenn Miller’s famous hit In the Mood. There are unsettled questions about the authorship of the three routines that played these recorded melodies. In the wake of Strachey's tour de force a number of people in the lab started writing music programs: even the routine that played the National Anthem in the recording may have been a retouched version of Strachey's original.
It was a challenge to write routines that would keep the computer tolerably in tune, since the Mark II could only approximate the true pitch of many notes: for instance the true pitch of G3 is 196 Hertz but the closest frequency that the Mark II could generate was well off the note at 198.41 Hertz. We found there was enough information in Turing's wonderfully pithy Programmers' Handbook to enable us to calculate all the audible frequencies that the Mark II could produce. However, when we ran a frequency analysis of the 1951 BBC recording (using the British Library's digital preservation copy, tape ref. H3942) we found that the frequencies were shifted. The effect of these shifts is so severe that the sounds in the recording often bear only a very loose relationship to the sounds that the computer would have actually produced. So distant was the recording from the original that many of the recorded frequencies were actually ones that it was impossible for the Mark II to play.
Naturally we wished to uncover the true sound of the computer. These 'impossible pitches' in the recording proved to be the key to doing so: our computer-assisted analysis of the differences in frequency—between the impossible pitches and the actual pitches that the computer would have played—revealed that the recorded music was in fact playing at an incorrect speed. This was most likely the result of the mobile recorder's turntable running too fast while the acetate disc was being cut: achieving speed constancy was always a problem with the BBC's standard mobile recording equipment at that time.7 So when the disc was played back at the standard speed of 78 rpm, the frequencies were systematically shifted.
We were able to calculate exactly how much the recording had to be speeded up in order to reproduce the original sound of the computer.8 We also filtered out extraneous noise from the recording; and using pitch-correction software we removed the effects of a troublesome wobble in the speed of the recording (most likely introduced by the disc-cutting process). It was a beautiful moment when we first heard the true sound of Turing's computer.
Here is the complete recording of our restoration:
1 See, for example, Chadabe, J. 'The Electronic Century, Part III: Computers and Analog Synthesizers', Electronic Musician, 2001, www.emusician.com/tutorials/electronic_century3.
2 Australian composer Paul Doornbusch writing in R. T. Dean, ed., The Oxford Handbook of Computer Music, Oxford University Press, 2009; see pp. 558, 584.
3 A. M. Turing, Programmers' Handbook for Manchester Electronic Computer Mark II, Computing Machine Laboratory, University of Manchester (no date, circa 1950); a digital facsimile is in The Turing Archive for the History of Computing, www.AlanTuring.net/programmers_handbook. Turing's Mark I/Mark II terminology was eventually superseded when the engineering company that was contracted to build and market the Mark II, Ferranti, called it the Ferranti Mark I.
4 Christopher Strachey interviewed by Nancy Foy in 'The Word Games of the Night Bird', Computing Europe, 15 August 1974, pp. 10-11.
5 Strachey in 'The Word Games of the Night Bird', p. 11.
6 Letter from M. H. A. Newman to Strachey, 2 October 1951 (in the Christopher Strachey Papers, Bodleian Library, Oxford, folder A39).
7 BBC Recording Training Manual, British Broadcasting Corporation, 1950.
8 We describe in detail how we did this in our article 'Turing and the history of computer music', in J. Floyd and A. Bokulich, eds, Philosophical Explorations of the Legacy of Alan Turing, Boston Studies in the Philosophy and History of Science, Springer Verlag, 2017.
Jack Copeland is Distinguished Professor in Arts at the University of Canterbury, New Zealand. His recent biography Turing, Pioneer of the Information Age contains more information about Strachey and the Manchester computer music (Oxford University Press, paperback edn. 2014).
Jason Long is a New Zealand composer and performer, focusing on musical robotics and electro-acoustic music. He has carried out musical research at the University of Canterbury, the Victoria University of Wellington, Tokyo University of the Arts, and the Utrecht Higher School of the Arts.