A voice on paper: the 1860 recording heard in 2008

The voice survived as a line. Scott had no way to play it back. When researchers finally did, another problem appeared: whose voice was it?

By Coru · Published 2026-10-25 · Updated 2026-10-25

The drawer contains a voice

Imagine opening a drawer and finding a voice you cannot hear. There is no disc to put on a turntable, only a wavering line on a sheet. You can follow it with your eyes. What would you need to turn that line into sound?

On 9 April 1860, Édouard-Léon Scott de Martinville recorded the French song Au clair de la lune. The recording was heard in 2008. The gap is 148 years, but the stranger detail is that Scott had not designed his recordings for playback.

A machine that wrote sound

The Smithsonian describes Scott’s phonautograph, patented in 1857, as an attempt to make sound visible. Sound moved a membrane; a stylus attached to it traced a line through soot on paper or glass. The result was called a phonautogram.

Think of the difference between a score and a recording. A score tells a performer what to play. This line was a physical trace left by an actual sound. Yet preserving that trace did not give Scott a way to listen to it.

The line gets a second life

In his account for the Library of Congress, researcher David Giovannoni describes locating and digitally preserving Scott’s recordings in French collections between 2007 and 2009. His team played the April 1860 song in 2008.

The recovery used images of the traces. First Sounds describes an initial method developed at Lawrence Berkeley National Laboratory: software followed the wavering line as a virtual stylus. A sheet could become a source of sound without being a playable disc.

Then the singer changed

First Sounds initially chose a speed at which the singer seemed to be a woman or an adolescent. In early 2009, another recording, a recitation from Aminta, prompted a rethink. A note in which Scott acknowledged his own verbal mistake helped identify that speaker.

A tuning fork had left a separate calibration trace, giving the researchers a clock for the unevenly turned cylinder. The researchers revised their reading of the timing reference from 500 to 250 vibrations per second. At the corrected, slower speed, the song’s voice resembled the Aminta speaker. First Sounds considers Scott a likely singer, but does not claim certainty.

A small clock experiment

Here is an invented example to make the timing problem tangible. Draw twenty equal spaces on paper and imagine that each holds one second of a recording. Moving across one space per second takes twenty seconds. Crossing two per second takes ten.

For ordinary playback without pitch correction, that faster journey also raises the pitch: each vibration repeats twice as often. Slowing it back down changes both duration and pitch. This is a simplified model, not a reconstruction of the archive’s restoration.

Before reading an identity into a voice, then, ask what clock you are using. The trace is evidence; the time scale is part of how you interpret it.

What does “oldest” mean here?

The current First Sounds catalog identifies the April 1860 song as the earliest recovered human-voice recording that is clearly recognizable. It also lists earlier voice traces from 1857 and a recognizable tuning-fork recording from 1859. “Oldest sound” hides those different categories.

For an archival listen, use the first source link below: the April 9 recording in the May 2010 restoration. The earlier-playbacks link preserves the speed revisions. These are archival restorations, not a modern reenactment. You can follow the whole explanation without listening.

A useful question to carry into any historical recording is: what belongs to the original trace, and what depends on the way we play it? In this case, even the apparent singer depended on that distinction.