Retrieving Sound from Soot
—Our Evolving Approaches
First Sounds pioneered the playback of phonautograms in 2007 and has been in the forefront ever since. Phonautograms were recorded to be studied visually—not to be played back—and the challenges of interpreting their waveforms are significant.
The Amplitude Domain
Between December 2007 and March 2008, David Giovannoni and Earl Cornell pioneered the methods that first made phonautograms audible. Giovannoni traveled extensively to digitize the original documents in high optical resolution, after which he prepared images for eduction. Cornell converted Giovannoni's images to sound using software developed at Lawrence Berkeley National Laboratory and modified for this application. Giovannoni assembled the audio files and undertook signal processing with Richard Martin, while Patrick Feaster manually corrected the time signal (below). The heralded March 2008 release of "Au Clair de la Lune" from April 9, 1860 was a direct result of their collaboration. (Giovannoni recounted this rapid development and deployment in "The Phonautogram Diaries".)
However, most phonautograms were not recorded with an understanding of the requirements of playback, and from a modern perspective their tracings are often malformed: the recording stylus left the paper, smeared its traces, marked several places at once, moved backwards along the time axis, and violated other fundamental requirements of signal recording. Berkeley's technologies failed when contronted by these conditions, and we initially feared that many phonautograms—particularly the earliest ones—might remain permanently mute.
Patrick Feaster devised an alternate playback approach in late 2008. Unlike the initial method, which treats a phonautographic trace as if it was a groove in a phonograph record, Feaster's approach graphically converts the trace into a signal of varying width that can be read as an optical film soundtrack. This approach cannot correct serious malformations, but it has proven sufficiently robust to let us hear something from phonautograms that are otherwise too compromised to process. All of the phonautographic sound files released by First Sounds since mid-2009 have been educed by Feaster using this approach.
The Time Domain
Édouard-Léon Scott de Martinville's most technically adept phonautograms contain two recordings made simultaneously—one from the membrane driven by airborne sounds, the other from a tuning fork (diapason) vibrating at a constant pitch. Scott's tuning fork presaged the 20th century's pilot tone and time code solutions to signal stabilization. Not only does it tell us the exact speed at which the airborne signal was recorded, it also provides extraordinary stability in the signal's playback.
Patrick Feaster manually corrected "Au Clair de la Lune" for its March 2008 release. Earl Cornell quickly wrote software to make the adjustments, and David Giovannoni premiered a revised "Au Clair de la Lune" at the Association of Moving Image Archivists' "Reel Thing" in June. At that symposium Bob Heiber (Chace Audio) and Bob Simmons (Sony) suggested working with Jamie Howarth of Plangent Processes. Howarth claimed Plangent's Clarity software gave Scott's phonautograph the stability of a Studer A80. The results were audibly superior and we re-re-released "Au Clair de la Lune" with this improvement. Our first eduction of “Gamme de la Voix” from May 17, 1860, which Giovannoni premiered at a keynote address at the Audie Engineering Society's convention in fall 2008, also benefited from Howarth's contribution.
Today, with Berkeley and Plangent no longer able to contribute their services, we are now exploring with Mathis Nitschke at Celemony how Capstan's time correction algorithms might be applied to phonautograms.
First Sounds has digitally preserved every airborne sound recording known to exist from before 1861. We aspire to hear them all, but to do so we must build our own software to adequately address all challenges inherent in phonautogram eduction.
Feaster and Giovannoni have drafted initial specifications for such software and actively seek a motivated and capable partner to write the code. Whether a software firm or an individual, that partner will have the unique opportunity to return the world's first recorded sounds to the world.
Email us at email@example.com for more information.