While I encourage the reader to browse the links above, I shall give a condensed version of the story here.
In 1857, Frenchman Édouard-Léon Scott de Martinville patented the phonautograph. The device used a stylus, connected to a receiving horn or bell, to inscribe a visual representation of the sound on to a recording medium (originally, a lamp-blackened glass plate; later, blackened paper). The purpose was to create a visual representation of sound -- not to provide for reproduction of that sound. And that is the key point to this story.
In April of 1860, a 10-second recording was made of "Au Claire de la Lune". This recording was recently found in the archives of the Académie des Sciences of the Institut de France.
Scientists working at the Lawrence Berkeley National Laboratory in California created a method to reproduce the originally-recorded sounds from their sooty spectra. In short, an optical scan of the recording paper was made, and then computer software decoded the transcription to reproduce the sounds which originally drove the recording stylus.
The scientists are part of a collaborative, First Sounds. According to their web site:
First Sounds is an informal collaborative of audio historians, recording engineers, sound archivists, scientists, other individuals, and organizations who aim to make mankind's earliest sound recordings available to all people for all time.
Photos by Isabelle Trocheris
So: these scientists took a visual representation of sound, which was never meant to be played back, and turned it back into sound.
Go back, and re-read that last sentence, considering the technology involved, and I think you would agree that this work is remarkable.
Are there more inadvertently-saved sounds waiting to be re-discovered? All that is needed, in theory, to record a sound is (1) a moving medium, and (2) a stylus which reacts to sound waves.
As is often the case, I am not the first to have such a fanciful notion. An episode of a popular American television science-fiction television series from the 1990s pondered what would happen if, say, sounds from 2000 years ago were recorded into a wet clay pot as a stick was being used to carve decorative grooves in it.
A little more research reveals that the concept was also the heart of an April Fools Day prank by a certain Bilge Sehir a few years ago, as nicely described on the Language Log blog. I mean, really: a 6500-year-old recording? Too good to be true.
Still, it does make for a nice gedanken experiment.
For quality sound reproduction, stick with Mr Edison's cylinders ... and beware of pseudoarchaeologists and/or April Fools pranksters.