Discover more from The Honest Broker
How an IBM Computer Learned to Sing (1961)
The IBM 7094 anticipated the future of music—and also sounded like the Auto-Tuned pop stars of today
Not many people could afford an IBM 7094 computer back in the early 1960s—a typical installation cost $3 million. That’s the equivalent of around $20 million in purchasing power today. Over the course of the decade, fewer than 300 were built.
You didn’t get much computing power for that hefty price tag, at least by current-day standards. But if you wanted a machine that did complex or rapid math, you had few other options. The 7094 could handle 250,000 additions or subtractions in just one second. A whole room of accountants couldn’t keep up with it.
But addition and subtraction aren’t very sexy. So someone got the bright idea of teaching the IBM 7094 to sing. That’s why John L. Kelly Jr., Carol Lockbaum, and Lou Gerstman of Bell Labs, in Murray Hill, New Jersey, began working in 1961 on this pioneering computer music project.
The Honest Broker is a reader-supported guide to music, books, media & culture. Both free and paid subscriptions are available. If you want to support my work, the best way is by taking out a paid subscription.
Digital music wasn’t an entirely new development, even in those distant days, but singing presented completely different challenges, requiring breakthroughs in speech synthesis. But Bell Labs—then the in-house research arm of AT&T (it’s now part of Nokia)—had more expertise in that area than any other organization in the world.
The Bell Labs team needed a song for their experiment. They decided on “Daisy Bell”—also known as “Bicycle Built for Two”—composed by British tunesmith Harry Dacre in 1892.
The idea for the song came to Dacre when he visited the US and found, to his surprise, that the customs officials had imposed a tariff on his bicycle. A friend quipped that he was lucky it wasn’t a bicycle with two seats, or the duty might have been double. The end result was Dacre’s most successful song ever.
Here are the lyrics:
Give me your answer, do!
I'm half crazy,
All for the love of you!
It won't be a stylish marriage,
I can't afford a carriage,
But you'll look sweet on the seat
Of a bicycle built for two!
They are called tandem bikes nowadays, or so I’m told. Not very popular—whether for traveling or courtship—but still available for purchase.
Even back in the early 1960s, this tune didn’t have much hipness potential. But at least the melody was simple, well-known, and no longer protected by copyright. (That said, I would love to watch a jury in 1961 debate computer music rights.)
“I’m especially fascinated by the decision to teach the computer lyrics that are essentially a marriage proposal….When the first AI chatbot proposes marriage in the year 2025, the happy couple ought to play this song at their wedding.”
For the instrumental parts of the song, the Bell Labs team relied on contributions from Max Matthews, who had created a breakthrough sound-generating program called MUSIC back in 1957. In those ancient analog days, he had hooked up his violin to an IBM 704, and was thus the first performer in history to transfer live music to a computer for synthesis and playback.
The Bell Labs team now built on this foundation with its new speech synthesis technology. The task must have seemed frivolous to many back then—who could have envisioned the future dominance of digital music at the dawn of the 1960s?—but the team drew on the talents of serious high-level scientists.
Gerstman, for example, would later emerge as one of the first experts in analyzing ‘voice prints’ and served as expert witness in the celebrated 1973 criminal prosecution of New Orleans D.A. Jim Garrison for bribery, where his analysis of a tape recording played a key part in securing an acquittal. Later he would turn his attention to helping people recover their speech skills after a stroke. But back in 1961 he was focused on the more lighthearted project of giving singing lessons to an IBM computer.
The entire performance of “Daisy Bell” lasted less than two minutes, with the vocal only featured for 30 seconds. It sounds creepy, but also surprisingly futuristic. Listening to the singing IBM of 1961, I can’t help but be reminded of current-day pop songs with a little bit too much Auto-Tune in the vocal. But for listeners back then, the closest equivalent might have been the corny megaphone fad of the 1920s, which found Rudy Vallée and others using this crude means of amplification on records and live performance. (Here’s another, more recent example.)
There are many curious aspects of this short snippet of music. I’m especially fascinated by the decision to teach the computer lyrics that are essentially a marriage proposal. And the happy-ever-after envisioned by the singer is seeing the beloved embedded in a new technology (albeit a two-seat bike, in this instance).
All this strikes me as a little edgy and transgressive, but perhaps I’m reading more into the situation than it deserves. But given the current-day investments in robots designed to provide relationships, broadly defined, you have to give the Bell Labs team credit for anticipating the future in this regard as well. When the first AI chatbot proposes marriage in the year 2025, the happy couple ought to play this song at their wedding.
This laboratory experiment had a strange sequel that continues to reverberate in popular culture. In 1962, science fiction writer Arthur C. Clarke visited John Kelly at Bell Labs, where he heard a demonstration of the singing computer. He was so struck by the performance that he incorporated it into the story line of 2001: A Space Odyssey, and it evolved into one of the most memorable scenes in cinema history.
When astronaut Dave Bowman needs to shut down the dangerous HAL 9000 computer, the machine’s abilities start to degrade. It regresses back to its earliest programming experiences, which included learning to sing “Daisy Bell.” Even today, that moment in the film is riveting, and all the more so when viewers realize that it references a real moment in the evolution of computer technology.
Perhaps the only disappointment here is that film director Stanley Kubrick rejected the idea of using an actual IBM computer for the voice synthesis in his movie. Instead he hired an actor. He picked Douglas Rain after hearing his narration in a documentary. He felt that this Shakespearean actor had the right stuff to provide “the creepy voice of HAL.”
How strange that an actor was better at emulating a computer, back in the 1960s, than a computer itself. We’ve come a long way since then. Today almost every aspect of music-making, from composition to curation, is getting handed off to machines. But 60 years ago, just teaching a computer to sing for 30 seconds was a technological marvel.
I just hope that AI-controlled music has happier results in real life than it did in that Stanley Kubrick movie. Given the pace of recent developments, we may find out soon enough.