Skip to main content
No. 2355:
A Picture of Music

Today, a picture of music. The University of Houston's College of Engineering presents this series about the machines that make our civilization run, and the people whose ingenuity created them.

You may've heard about a newly-discovered recording of Au Clair de la Lune -- made in 1860, just before our America Civil War. It's being touted as having priority over Edison's cylinder phonograph. Before we hear this old recording, remember what the French tune sounds like: 

[audio of the first words of Au Clair de la lune.]

Now listen to this barely audible newly found version: 

[audio of the phonautograph recording].

We've long thought that Edison's most original creation was his phonograph. At the same time, no priority claim is ever entirely valid. We can trace antecedents for most of Edison's inventions. But his phonograph seemed to stand alone until recently. Then other claims began surfacing.

phonautographThis one was made by Éduard-Léon Scott de Martinville. Starting in 1853, he'd worked on a device that would make a picture of a sound. Someone spoke or sang into a large barrel. At the bottom was a vibrating stylus. The stylus scratched a continuous track on a paper sleeve, coated in soot, as it turned on a rotating cylinder. Those sleeves are what remain. 

De Martinville believed that seeing sound would help us understand it. And, with that in mind, he kept improving his so-called phonautograph.The version he patented gave us the picture from which this ghostly voice emerges.

de MartinvilleA group headed by David Giovannoni at Stanford located the old de Martinville paper sleeves and began the complex process of extracting audible sound from them. They used a technique developed in collaboration between Berkeley's Lawrence Laboratory and the Library of Congress. The earlier paper sleeves gave only static, but this one finally yielded a barely audible song.

The catch is, de Martinville never meant to produce sound. He would be as amazed as we are by this old sound bite. Edison's whole aim was to play a captured sound back to us. But here's a strange coincidence: De Martinville chose the most basic children's song. And what did Edison do for his first recording? Being no singer, he merely recited Mary Had a Little Lamb. That's the song where American children still begin their musical study. 

Both tunes start in the simple key of C and are largely sung or played on the white notes C, D, E. (I should point out that de Martinville's singer emerges a whole step higher.) 

But we're left with questions of preservation, more than invention. A trained musician often looks at notes on a page and hears them in her head. Notes are just another means for encoding sound. So I offer a challenge to you techies: I'll sing and record this same bit of Au Clair de la Lune. Then, in the web version of this episode, I'll post the picture of the sound from the recording screen. If you can reclaim a sound from that picture, send it to me in an mpeg file -- and I'll post it for all to hear. 

I'm John Lienhard, at the University of Houston, where we're interested in the way inventive minds work.

(Theme music)

De Martinville's phonautograph has received wide coverage on the web. See also, J. Rosen, Researchers Play Tune Recorded Before Edison. New York Times, Thu., Mar 27, 2008. Images above are courtesy of Wikipedia.

Let me point out a confusion factor that arises here. One will find reports of an earlier (1848) recording of Chopin playing his Minute Waltz. Such a recording was reported in 1980 and later debunked as a fraud. (My thanks to Chris Hathaway and Dayton Smith at KUHF for chasing this one down for me.)

For the visual audio challenge mentioned above Click Here. We are genuinely curious as to whether anyone will find means for obtaining sound from this visual record of a four-second sound bite. (These images were obtained using the Kool Edit recording system.)

Note added on April 12, 2008. Only one person has responded to the challenge to date. Tim Warburton, Department of Computational & Applied Math at Rice University, extracted the signal from my photos of the sound trace. The quality is only a little better than the deMartinville sound. The reason is that the information in the eight traces that I provided hardly improves on that in deMartinville's paper sleeves. I would have to have provided a vastly expanded record -- maybe ten times as many photos, to get a decent sound. Many thanks, Dr. Warburton. 

Note added on April 25, 2008. Podcast listener and electrical engineer Tom Divine heard this episode a while after it aired an provided his own response to the challenge. He spliced and cropped the images, cleaned them up and converted the result to a bitmap file. He used his own code to take a weighted average of the pixel brightnesses in each column to estimate the y-coordinate of each sample, and wrote the results to an digital audio file. 

Note added on Aug. 17, 2010. Kurt Nauck of Nauck's Vintage Records in Spring, TX, writes to point out some very interesting codicils on this episode. First of all, David Giovannoni only premiered his work at Stanford. He's actually a member of a group call First Sounds. Since then, he and one Patrick Feaster have done further work. Now, it seems, the voice you hear is actually that of de Martinville himself, not a woman. They've also gone much further with their work. For more on the matter, see the First Sounds website, here.