On a winter afternoon in early 2000, Carl Haber, a bearded man in his mid-40s, may have looked like any other collector as he walked into Down Home, a music store in El Cerrito, Calif., picked out a stack of 78-r.p.m. records, paid for them, and drove away.
But Haber is not a record collector. He is an experimental physicist. And rather than placing the 78s on a turntable, he did something that had surely never been done before in the long and warmly crackling history of the phonograph. He took them to his office at the Lawrence Berkeley National Laboratory and placed them under a pristine $10,000 Leica stereo zoom microscope.
What he saw delighted him, for one simple reason: There were clear and — critical to what he had in mind — measurable edges to the grooves.
A few weeks earlier, Haber had been casually listening to NPR when a story came on about the millions of early sound recordings that are deteriorating in the vaults of the Library of Congress and elsewhere. He learned that it was a race against time to digitize these recordings before they become unplayable. Some of them are so delicate they can no longer be touched with a traditional stylus.
“There was this ‘aha’ moment,” Haber recently recalled. “I realized here is a problem for physicists to try to solve.”
More than a decade later, Haber has won wide recognition, and a MacArthur Fellowship, for a revolutionary image-scanning technology that has the power to pull sound from rare and fragile recordings without touching them, and in so doing, to help protect some of the most vulnerable corners of this country’s aural heritage.
For years this technology was put to use only in Haber’s lab and at the Library of Congress, but over the last few months, a large converted mill building in Andover has become home to the fourth groove-scanning system in the country.
On a record, grooves are the channels into which sound waves were once etched. What if, Haber wondered, he could create a precise image of these grooves with enough detail to play back the sound not from the grooves but from the image itself?
Haber explored this idea on and off for the next three years. A few of his colleagues joined in, looking at 78s under microscopes whenever they had free time between experiments.
Others in the lab considered the whole enterprise rather quixotic — “a joke,” in Haber’s words. But in fall 2002, a colleague named Vitaliy Fadyev played Haber a snatch of audio he had finally succeeded in lifting from images captured by the lab’s special cameras. The music was from a recording of “Goodnight, Irene,” sung by the Weavers. The idea had worked.
“Carl is a hero,” said Gene DeAnna, head of the recorded sound section of the Library of Congress. “If we’re going to succeed in preservation we need science. And we are absolutely in awe of what he’s been able to do.”
Haber and his colleagues named their technology Image Reconstruct Erase Noise Etcetera, but nobody calls it that. Instead they use the acronym that was coined in tribute to that first 78: IRENE.
The IRENE system in Andover lives at the Northeast Document Conservation Center, an independent conservation lab known for its work on paper-based collections. In a small back room, a sound archivist has been working daily over the last few months next to two of IRENE’s cameras mounted on a flat surface about the size of a ping-pong table, scanning all varieties of records as well as older wax cylinders whose grooves are etched vertically and therefore require a special 3-D scanning technique.
On the wall of the room is a poster-sized enlargement of “When Lilacs Last in the Dooryard Bloom’d” — not the poem by Walt Whitman but the song by Paul Hindemith. From a distance the image looks like an abstract striped pattern, with jagged black lines standing out against a white background. These are in fact the grooves of a recording of the CBS Symphony Orchestra performing on June 30, 1946. Look at them closely, and you are seeing sound.
A look into early recordings
Over the last decade, scientists, audio historians, and museum curators around the country have together harnessed IRENE to make a slew of sensational discoveries. Among them was the first playback of one of the world’s earliest recordings of a human voice: a Frenchman singing “Au Claire de la Lune” in 1860, almost two decades before Edison invented the phonograph.
The recording lasts just a few hair-raising seconds, but the singer’s voice, while muffled, is eerily present. The surface noise has the effect of conveying a sense of vast temporal distance. We are listening in at the dawn of recorded sound.
IRENE has also opened a remarkable window onto the earliest American experiments in sound recording, by extracting sound from discs — made from tinfoil, photosensitive glass, or wax-and-cardboard — used at the Volta laboratories in Washington, D.C., the research home of Alexander Graham Bell. In the process IRENE has given us the only known example of Bell’s voice, recorded in 1885, stating with clarity and audible pride: “Hear my voice: Alexander Graham Bell.”
Then there are the countless one-of-a-kind field recordings, often documenting languages no longer spoken or folk rituals long since vanished. “Those are all incredibly significant materials that are physically at the end of their lives,” said DeAnna. He estimates that the window to preserve these recordings through digitization is 20 years, though other experts say we have even less time.
Haber’s team has worked with the Phoebe A. Hearst Museum of Anthropology to scan fragile cylinder recordings that were made of a Native American named Ishi who was thought to be the last member of California’s Yahi tribe, and it has scanned selections from Harvard’s Milman Parry Collection, a fascinating trove of field recordings made in Yugoslavia in the 1930s by a maverick classics professor studying the transmission of oral poetry.
Some 3,000 aluminum transcription discs are sitting undigitized at Harvard’s Widener library, featuring South Slavic song, recited poetry, and interviews with people whose lives stretched back in some cases well into the 19th century.
The Northeast Document Conservation Center is also tackling wax cylinders from the Helen Hartness Flanders Ballad Collection at Middlebury College, a repository of 4,800 field recordings of New England folk songs and folklore, captured in the homes and on the front porches of Vermonters over three decades beginning in 1930. One of the center’s first successes this year was a vivid rendering of a sprightly fiddle tune called “Up and Down the Grade,” recorded in Burlington, Vt., in 1931.
And that’s not the only folk music to be heard in recent weeks at the center. Back in 2010, the Woody Guthrie Archive received a donation of nearly two dozen 16-inch transcription discs labeled only with vague terms like “Jam Session #1,” but the discs were never played for fear of harming them. Many radio stations across the country recorded their programming on similarly fragile lacquer discs that were designed for only a few plays.
“We thought IRENE would be perfect for these discs,” said Nora Guthrie, Woody’s daughter, who contributed the mystery records to the current pilot study.
The conservation center has only recently begun working on these discs, and Guthrie received an e-mail with an IRENE-created digital sound file attached just before speaking with the Globe. “It sounds like my living room in the 1950s!” she said. “And there’s a voice and a banjo I think I recognize as belonging to a young Pete Seeger. Who knows, maybe there will be a new song on one of these discs that no one has ever heard.”
Assessing the technology
Yet just how widely the technology will ultimately be used remains to be seen. A set of crucial questions remains. Can IRENE — which Haber originally adapted from imaging techniques used in his particle physics research at the Large Hadron Collider at CERN in Switzerland — be deployed beyond the physics and research communities as a practical tool in the sound preservationist’s toolbox? Is it robust, scalable, and affordable enough to help save the millions of undigitized recordings that collectively represent humanity’s slowly crumbling aural heritage?
The preservationists at the conservation center have set out to address these very questions. The “Lilacs” image on the wall was captured from a record on loan from Carnegie Hall, which is one of several partner institutions participating in a pilot study to determine how IRENE can best be used. “In research, you can fund a thing forever, and you can always improve it, but eventually you’ve got to have a practical application,” said Bill Veillette, executive director of the conservation center. “That’s what we’re exploring right now.”
On a reporter’s recent trip to the lab, sound archivist Mason Vander Lugt demonstrated the technology by first playing an old commercially produced Edison wax cylinder recording of Arthur Collins singing his ragtime hit “The Preacher and the Bear,” through a vintage cylinder player from the same era.
The sound had the expected hisses and pops but was relatively clear. Then he turned to a pair of large computer monitors, where one could see the images that IRENE had created of the same cylinder through a process of 3-D scanning. White and gray stripes represent the hills and valleys of the cylinder’s grooves, but there was something else on the picture, a smattering of inky blotches that looked like weeds overtaking the rows of a neatly organized garden. This, in fact, is mold, the preservationist’s scourge, since it alters the wave form of the sound and slowly eats away at the wax cylinder.
Lugt can clean up the images to some extent the way one might touch up a photo with a program like Photoshop. He could even conceivably pull the sound off of the pieces of a broken cylinder or disc and then stitch multiple images together. But creating the image is only part of the challenge. A separate software program reads the sound from the picture.
And yet when Lugt played back the sound from even a well-cleaned IRENE image, it was still full of hiss. We were hearing, he explained, everything on this image, so the music can sometimes sound significantly recessed behind the layers of surface noise.
This is where traditional audio engineers can step in and work their magic on the sound file, the way they might process any other digital file of historic audio. IRENE’s own native sound-rendering software will also continue to improve over time. And with particularly rare or fragile media, IRENE’s non-contact approach may be the only option. The key prize in all of this, said Tom Rieger, director of imaging services at the conservation center, may not even be the audio but the image itself.
“IRENE is a game-changer in that it creates a perfect record of the groove structure,” he explains. “All other playback systems, either with a stylus or a laser, are an interpretation of the groove. With IRENE the image of the audio becomes the preservation master, not the actual audio, which we would consider to be derivative.”
Buying preservation time
In other words, IRENE has the potential to capture old recordings in their present state before they deteriorate any further. If nothing else, that can buy preservationists time, which, as it turns out, they may need more than anything else.
To get a sense of the potential scope of the challenge, consider the Library of Congress’s Packard Campus for Audio Visual Conservation. Completed in 2007 on a 45-acre site near Culpeper, Va., it is the largest archival facility of its type in the world, with nearly 120 linear miles of shelving.
Older videocassettes are being robotically digitized there 24 hours a day, and a host of sound transfer technologies are being deployed to handle audio recordings. But even at this state-of-the-art facility, the problem is only growing graver. According to DeAnna, the facility digitizes about 15,000 recordings a year, but it is acquiring 250,000 a year. “That disparity is what concerns me,” he said, because the longer some recordings wait, the further they deteriorate.
And the Library of Congress is hardly the only institution grappling with this problem. DeAnna estimates that there are tens of millions of recordings held at archives across the country. “In general, institutions that hold sound recordings don’t really know what to do with them. They very often get neglected and stashed in the vaults.”
Many of these recordings may still be most efficiently transferred using stylus-based technology. IRENE’s main utility could be in dealing with the hardest cases.
“IRENE is fantastic because it allows us to hear recordings that we otherwise would not be able to hear,” said David Giovannoni, an independent audio historian who is a member of First Sounds, the research collective that first tracked down the 1860 recording. “The downside is cost, but the good news is that we don’t need this new technology to transfer and preserve 99.99 percent of our grooved audio heritage.”
Haber speaks modestly about IRENE’s potential reach.
“I’m not trying to argue that this should take over the world and be the only way that people transfer things,” he said. “There are some collections where you wouldn’t use this technology because you could do a better job with standard methods. But I’m very pleased if the interest in early recordings that this work may have helped spark facilitates a general recognition and support for all of these amazing collections.”
In the end, scientists and preservationists are at least now joined in working toward the same goal of pushing back against the approaching ocean of silence. Or as Veillette puts it, “everything we’re doing is to make sure this stuff is around to be heard 100 or 200 years from now.” He looked out over the stacked boxes of recordings that were still waiting to be scanned. “The ultimate audience is posterity.”