A race to restore the voices of the past
Connecting state and local government leaders
Energy Department physicists have developed an imaging system that allows libraries and museums to recover voices from audio recordings that have been mute for more than a century.
In the 1880s, Alexander Graham Bell and his associates working in Washington performed some of the earliest experiments with optical transmission and sound recording. Nearly 130 years later, a team of physicists, curators and preservationists is using high-resolution digital imaging to tease the sound out of these and a handful of other experimental recordings.
Six recordings, created by Volta Laboratory Associates between 1880 and 1885, are among nearly 200 recordings Bell deposited with the Smithsonian Institution. They were played back in 2011 using IRENE (for Image Reconstruct, Erase Noise, Etc.), an imaging workstation developed by physicists at the Energy Department’s Lawrence Berkeley National Laboratory and programmed to interpret the imaged grooves.
The results are not exactly broadcast quality, said Carlene Stephens, a curator at the Smithsonian’s National Museum of American History. “You have to suspend your 21st-century sensibilities of what is good-quality sound,” she said. But considering that these recordings were made using a variety of techniques on media, including glass, copper, brass and wax and that no playback equipment was ever created for them, the recovery is remarkable.
Related coverage:
IRENE: Key to unlocking mute recordings
The Library of Congress now has two IRENEs, a 2-D and a 3-D version, and is using them to preserve recordings from the library’s collection that are in danger of being lost or becoming unable to be readily played.
The Library of Congress is the largest library in the world, with more than 147 million items in its collections, and preserving and accessing these recordings is a major challenge.
“The bulk of our collections are not books,” said Dianne Van der Reyden, the library's director for preservation. The library’s Packard Campus for Audio-Visual Conservation in Culpepper, Va., is home to more than 4 million items, including millions of analog audio recordings in a large variety of formats and media and in varying conditions of preservation.
“Few people are researching what to do with this material,” Van der Reyden said. Some media, such as shellac discs, are relatively hardy and can last nearly forever. But some recordings lack playback equipment, some are too fragile to be played, and some are physically deteriorating.
“This is a ticking time bomb," she said. "We’re in danger of losing much of our culture.”
High-energy physics and sound preservation
Because of the danger of losing some recordings forever, “we have really embraced digital preservation,” said Peter Alyea, the library’s digital conversion specialist. Reformatting old analog recordings to an archival standard can not only preserve them for the foreseeable future but also make them easily available for listening without additional wear and tear to the original. But some way is needed to recover sound from obsolete, damaged or fragile media without damaging them.
Carl Haber, a scientist at Lawrence Berkeley, became intrigued by this challenge in 2000. “I’m a physicist,” Haber said. “I work in high-energy physics, and my particular area of interest and expertise is instrumentation” for data collection. He and fellow scientist Earl Cornell put their heads together, and “we saw immediately that there was some relevance of the techniques we were using” to audio conservation.
After some brainstorming and “Saturday experimenting” on the concept, Haber and Cornell came up with promising results and approached the Library of Congress. The library and several other institutions contributed some funding, and Haber, Cornell and the occasional grad student spent the next several years developing IRENE. Additional support came from DOE, the National Archives and Records Administration, the University of California, the Institute of Museum and Library Services, the National Endowment for the Humanities, the Andrew P. Mellon Foundation, and the John Simon Guggenheim Memorial Foundation.
Conceptually, the idea was simple: Use a scanner to produce a high-resolution digital image of the grooves in a record, cylinder or other recording media, showing the details in three dimensions. Then, clean the images up to compensate for imperfections, wear, damage or errors in the imaging.
“With that information, we have algorithms that can calculate how the needle would move through that,” Haber said. Those virtual movements then can be used to duplicate the sounds that would be produced by a real needle or stylus.
The evolution of IRENE
“The first demonstration was pretty easy,” Haber said. “It followed fairly directly.” But the devil was in the details, and making it work well on objects of different shapes with recordings in different formats without constantly having to tweak the software was more complex. Although it is not yet as user-friendly as it could be, IRENE has become a versatile tool.
“As long as there is something like a groove, we have parameters built into it to adjust for the basic things that characterize it, such as size, depth, et cetera,” Haber said.
By 2006, the first 2-D IRENE was installed at the Library of Congress for production use. The 3-D version was installed in 2009. It is capable of producing images of grooves in three dimensions, providing more information on the depth and height of grooves. IRENE has now been used to digitally read and copy hundreds of rare recordings, preserving them digitally while gathering information on the technology’s strengths and weaknesses and its possible uses.
“We are in the middle of a broad study of how well we can tune IRENE,” Alyea said. It has proved capable of accurately imaging and reproducing many types and shapes of recordings, and because it does this noninvasively it does not risk the damage that could be done by playing a disc or cylinder with a traditional stylus.
It is not perfect, however. “It doesn’t get quite the fidelity you would get out of a turntable and stylus,” Alyea said. So, for stable, robust recordings such as shellac discs and for more recent high-fidelity disc recordings from the 1960s onward, it probably makes more sense to continue playing them the traditional way.
Simple tech, complex art
For the time being, getting the best results out of IRENE remains something of an art. “In some ways it’s fairly simple technology, but it is also complex,” Alyea said. “Sometimes it works really well,” but wear and variations in different types of recordings can degrade results and require additional tuning or tweaking of the software.
The software is being refined so that it does not have to be tweaked and tuned for each type of recording and can be used more easily by people without technical expertise. Over the past year, the library has collected 4 terabytes of data with IRENE from recordings of many different formats and conditions.
One of the possible additional uses of IRENE is to analyze recordings before they are played with a stylus, so that technicians can tell in advance what the best shape and style of stylus would be. This currently is often determined by trial or error, which adds to the wear and tear on old recordings.
“We’ve had some successes and some failures,” Alyea said. “It certainly is getting better.”
One of IRENE’s successes came in 2008 when it was able to recover the contents from what is believed to be the first sound recording, made on smoked paper in France in the 1860s, well before Thomas Edison’s invention of the phonograph in 1877. This recording was made as an experiment to show that sound travels in waves and was never intended to be played back. Using two-dimensional imaging, IRENE was able to read the tracings on the paper and reproduce the sound.
When Stephens read of this, she thought of the 200 Volta Lab recordings locked away at the Smithsonian. “This is what I’ve been waiting for 35 years” to hear, she said.
Bell's Volta Lab
The Volta Laboratory was established in the Georgetown area of Washington by Alexander Graham Bell, his chemist cousin Chichester Bell, and Charles Sumner Tainter in 1880. Over the next several years, they experimented with the transmission and recording of sound. Using revenue from the lab, Bell was able in 1887 to found the Volta Bureau, an institute to aid people with speech and hearing disabilities.
The 1880s was a period of innovation in recording and intense competition among Bell, Edison and Emile Berliner. To document their work to support patent claims, these inventors deposited about 400 early recordings with the Smithsonian along with notes and other records of the experiments. Some of the documentation also is housed in the Library of Congress.
Some of the Volta experiments were commercially successful. The graphophone, which recorded on a wax-covered cylinder, become a popular business tool for dictation and eventually evolved into the Dictaphone.
“But most of the recordings in the collection predate any kind of commercially available recording or playback systems,” Stephens said. “They were mute artifacts,” well-cared-for but incapable of being played.
IRENE, with its noninvasive, format-agnostic approach, offered hope of unlocking the old recordings, and Stephens approached the Library of Congress about the project. The library collaborated with Haber on a pilot program to demonstrate whether it was possible to recover the audio.
“Nothing about it was easy,” Stephens said. “The challenge was to tune the equipment for the nonstandard formats.”
Deciphering the earliest recordings
One of the earliest recordings recovered was from a photographic glass disc made in 1884. Bell had experimented with different techniques to modulate a beam of light by width and intensity, which was recorded in a spiral on the disc. IRENE was able to treat the images like the grooves on a record. The disc’s label identified it as a man saying “barometer,” but it took a while to identify the sound because the man was saying the word one syllable at a time — “ba-ro-me-ter.”
A wax recording of Hamlet’s soliloquy on a brass cylinder was easier to identify, Stephens said. “Right up front you could hear ‘To be or not to be.’ ” A recording of "Mary Had a Little Lamb" was a little harder to make out but still possible to hear.
Six of the old Bell recordings have been played back in the feasibility study, and now Stephens would like to see a broader program to recover early recordings in the Smithsonian’s collection.
“I don’t think it is feasible to do all of them,” she said. “Some are too fragile or too partial to get sound from.” She estimated that about half of the 200 Volta recordings are good prospects for playback and some of the others are possibilities.
Some experts maintain that Bell made some recordings himself.
“We would love to get confirmation of a recording of Alexander Graham Bell’s voice,” Stephens said. There now are no known samples of it, so identifying it on a recording would not be easy. But the Library of Congress has the lab notes of Bell and his cousin Chichester, and the Smithsonian has Tainter’s notes. “It’s a matter of collating the information.