27 Sep 2016

Restoring the first recorded computer music

From Jesse Mulligan, 1–4pm, 1:24 pm on 27 September 2016
Distinguished Professor Jack Copeland (right) and composer Jason Long (left) have restored the earliest known recording of computer-generated music, created more than 65 years ago.

Distinguished Professor Jack Copeland (right) and composer Jason Long (left) have restored the earliest known recording of computer-generated music, created more than 65 years ago. Photo: University of Canterbury

The first ever computer-generated music was recorded 65 years ago by scientist and computing pioneer Alan Turing and has been restored by a University of Canterbury Professor and one of the university's alumni.

Professor Jack Copeland and composer Jason Long have used specific programming techniques to restore the earliest known recording of computer generated music. The original music was generated by a primeval computer in 1951 in Manchester.

Read an edited excerpt of the interview below:

How did you hear about the recording?

The recording was never lost, exactly. The BBC heard that the Manchester computer was playing music, so they sent an outside broadcast van out to Manchester with an interviewer and a portable disc cutter in the van, which actually cut the disc while the computer was playing.

So the BBC produced two discs in the end. One which they took away and presumably it was played on one of their programmes, but it’s kind of disappeared. Probably chucked out by the BBC because they never had room to store it all, unfortunately. They even had recordings of Turing’s voice, which they threw away.

The other disc that they cut in the same session was kept by one of the computer engineers and he knew he’d gotten something really important, so he treasured it until the end of his life and then he passed it on to another one of the Manchester engineers, who took it down to London to the National Sound Archives to get a digital preservation copy made. I knew that was there in the sound archives, so I went along to listen to it. So I sat in this tiny little booth listening to it on headphones, to this incredible music from the past. I was just hooked.

I wanted to do more, so I got a copy of it from the archives, brought it back to New Zealand and Jason Long, my collaborator, and I did a frequency analysis of what there was on the recording and we pretty soon found out that that was not how the computer had sounded.

We got hold of Turing’s programming manual for the computer, the world’s first computer programming manual and we worked out from that what notes the computer could play. What frequencies. The recording was full of frequencies that the computer couldn’t possibly have played. So something had gone wrong somewhere and you weren’t hearing the true sound of the computer in that recording.

So you had to set about working out what the original sound was. There are some voices in the recording as well, did the pitch of the voices give you a clue to the intended pitch of the computer?

No, not really. In the restored version the voices they certainly sound a bit more natural than they did in the raw version. It was really what we knew from Turing’s programming manual, what frequencies should be there and they weren’t. So what we think had happened, and this is going back 60 years, the actual disc cutter that the BBC had in their van, it had a rotating turntable and then a cutting head which gauged out a groove on the acetone disc, that’s how they made the recording while the computer was playing and these things were notoriously iffy and the turntable didn’t always rotate at the right speed and we think that that must have happened with this recording. The turntable was rotating a bit too fast and as a result the recording was running to slow. That’s what we discovered. For whatever reason, the recording that we got from the national sound archive was running too slow. So that’s our best guess, that the culprit was the BBC’s disc cutter.

So you sped it up until you matched frequencies that you knew the computer could actually create, is that right?

Yes, exactly. It’s like if you have a tape recording or something that’s playing too fast, everything sounds squeaky. If it’s playing too slow, everything seems to low and long drawn out, voices sound low and so on. It was only a few percent that it was too slow but we wrote a computer programme that worked on the original recording and it searched for the original speed. Once it had delivered that number to us, then we knew how much we had to speed up the recording in order to get it sounding right.

We wrote a programme that searched through all of the possible speeds that it could have been playing at. We started really low and went up higher and it just did a brute search through all of the possible speeds until it found the right one where every frequency in the recording fitted really closing with what the frequency should have been according to Turing’s programming manual. So then we just adjusted the speed of the recording and suddenly we heard the true sound of Turing’s computer. It was like on an archaeological dig seeing something beautiful and important coming out of the earth.