Project IRENE: Lawrence Berkeley National Laboratory, Library unite to save Native American voices

In the twinkle-lit cafes and sage-incensed dorms around town, vinyl is all the rage among the youths.

But in the unassuming Moffitt Library — still shy about its hipster cred — it’s all about wax cylinders.

Tucked away on Moffitt’s second floor, in the Digital Imaging Lab, a team of researchers are restoring a trove of wax cylinders: the original vinyl. The objects hiss and pop just like a record, but they also happen to contain the sacred songs and voices of Native Americans, recorded by field anthropologists over four decades, from 1900 to 1940.

In the imaging lab, Stephanie Battle slides two fingers into a small brown cylinder, lifting against its smooth center. She’s careful to avoid the surface, whose priceless grooves could melt slightly under her touch.

Battle is the digital imaging specialist for Project IRENE, a campuswide effort to scan and digitize nearly 3,000 wax cylinders held in UC Berkeley’s Phoebe A. Hearst Museum of Anthropology. There are over 77 Native languages from California represented in the collection, some of which have transformed or faded away.

And today, the shells preserving those voices have started to decay.

“These cylinders are degrading every day,” Battle says. “We have to take this moment to capture them.”

The objects are artifacts of Thomas Edison’s 1877 phonograph, the first device capable of audio playback. A grandfather to the record player, the machine translated soundwaves into movement. Vibrations sent a small cutter bobbing up and down, carving patterns into tinfoil (and later wax, in Edison’s sequel) that a stylus could retrace.

But that wax is terribly brittle. Just playing the cylinders would chisel off precious material.

In 2003, researchers at the Lawrence Berkeley National Lab discovered how to scan the cylinders using only light. Most of them have been successfully scanned; the hundreds that remain, though, are in rough shape, either cracked into pieces or sunken with mold.

So LBNL scientists Earl Cornell and Carl Haber went back to the drawing board. Now — with help from the Library’s Makerspace — the two are running a marathon of trial and error that would make Edison himself proud.

Their challenge: How do you scan a handful of shards?

Or, how in the world do you not?

Student prints a mandrel in Makerspace
Student Madison Gaborko talks about a 3D-printed mandrel she's making for Project IRENE in Moffitt Library in March 2018. (Photo by Jami Smith for the UC Berkeley Library)

Freedom to fail

It’s early morning, and Maddie Gaborko, a fourth-year English major, bustles into Moffitt’s Makerspace. She plops down her coffee and breakfast sandwich and hovers over a computer, waking it from sleep.

Over her shoulder, a robotic arm loudly purrs as it spins around, dropping plastic from its tip. Layer upon layer, it builds a long gray rod full of hundreds of holes.

Gaborko had started the 3-D print at 10:30 the night before, after receiving a design from Cornell, the software scientist. This is the third object Gaborko has printed for the Project IRENE research team, each iteration one step closer to perfection.

“When you 3-D print, you have to consider tolerance,” Gaborko says of the process. “The plastic holes expand, so you may design something very specific, but the shape is still probably going to be a bit bigger than what you planned.”

The small towers coming to life are mandrels — rods that fit inside the cylinders to rotate them before a scanning machine. As the mandrel turns, a high-resolution laser measures the landscape, helping focus the 3-D camera below.

Before Makerspace came into the picture, the team in Moffitt had been strapping the broken cylinder pieces together. They scanned them over and over, moving the straps for each scan and then stitching the images together. That wasted a lot of time and resources, Battle says.

The team could also send the broken cylinders to a professional conservator to glue them back together. That costs about a thousand dollars each, and even the slightest error could damage them further.

So Cornell and Haber went to work, tossing ideas around their office. Up at the Lawrence Berkeley Lab, in the hills overlooking campus, tools and prototypes are strewn around their work tables, with measurements and musings covering the walls.

Left to right: LBNL physicist Carl Haber scans a record; LBNL software scientist works on Project IRENE software. (Photos courtesy of Lawrence Berkeley National Laboratory)

“These are all of the toys,” Cornell says, smiling as he navigates the maze. “It’s kind of a mess because we’re always trying new things.”

The secret, the scientists figured, lay in the mandrel. At first, Haber had designed a mandrel that would hold a broken cylinder together with clamps, with holes at each end for screws. Staring at the prototype, though, Cornell had another idea.

“It struck me that, well, if those holes had vacuums, it would work,” he says.

With vacuum suction, the shards can be snapped to a mandrel and hold their shape during scanning.

Cornell had staff at his daughter’s boarding school 3-D print a pretty rough trial version. Then, at the suggestion of Erik Mitchell and Lynne Grigsby — former associate university librarian, and the head of Library IT, respectively — Cornell and Battle turned to the Makerspace.

Using a professional machinist, 3-D printing the mandrels could cost thousands of dollars and weeks of time — crippling factors when it comes to exploration and taking chances.

Enter Makerspace. The 3-D printers in Moffitt are available to the entire campus community.

“It’s very cool to see the looks on people’s faces when they ask, ‘So what do I have to pay for this?’ And I say, ‘It’s free, just come do it,’” Gaborko, the Makerspace worker, says.

3D-printed letters in MakerspaceMakerspace tools area
Top to bottom: From left: 3-D printed letters await pickup in Makerspace in Moffitt Library. Makerspace has 3-D printers, tools and spaces for building and creating. (Photos by Jami Smith for the UC Berkeley Library)

For now, Cornell is trying to solve what seems to be the last puzzle: creating a perfect seal for the vacuum suction, ensuring the broken cylinder pieces stay safely attached to the mandrel.

Normal 3-D-printing material is quite rigid, so a lot of the airflow is lost in the cracks. Tiny suction cups could be inserted into each hole to secure the seal, Cornell says, but they cost $2 a piece. Instead, Cornell is designing a new mandrel to print in the Makerspace — one made of special, springy material, with suction cups built in.

“I very likely may have to modify this one, too,” Cornell says, thinking out loud. “I’m not sure how big to make the (suction cup) wings that are flexible.”

The capabilities at Makerspace also allow the Project IRENE team to experiment with other kinds of materials, including the supersized cylinders designed by Edison to record long concerts. Cornell bought one of the giant cylinders on eBay, curious if he could design a mandrel to fit.

Back in the Digital Imaging Lab, Battle moves to a computer and selects a file: a 1901 recording of a whistling concert by famed “Birdman,” Joe Belmont, recorded on a concert cylinder. From the quiet hum of the lab bursts the sharp, allegro tune of a bird’s song.

“This was $50 bucks on eBay, because nobody was able to play it back,” Battle says. “But we can.”

For Battle, a musician with a background in archaeology and art history, the constant and careful adjustments carried out by students in Makerspace give researchers peace of mind. They can make sure each mandrel is perfect before trusting the machine to hold such valuable recordings.

“The best way to ensure the preservation of the object is to have something that is extremely unique to each object,” Battle says. “Makerspace is allowing us to do that.”

“It’s literally down the hall, and that’s great,” she says.

Stephanie Battle works with scanner
Stephanie Battle, digital imaging specialist for Project IRENE, sets up wax cylinders for scanning in Moffitt Library in April. (Photo by Jami Smith for the UC Berkeley Library)

Coding the cracks

After the cylinders are prepped and scanned, their images are fed to software created by Haber and Cornell. The software acts as a digital stylus of sorts, slipping into the waves on screen, singing as it goes.

But there are nicks, hair, and dust along the cylinders — not to mention small pits from what is believed to be bacteria that have nibbled on the cylinders for a century. During playback, those intruders sprinkle sharp thwacks and crackles throughout the cylinders’ songs.

With a few programming tricks, a team of undergraduate students are making many of those blemishes disappear. The students even help code and debug the software, Cornell says.

In Moffitt’s basement, Youmna Rabie, who is heading into her junior year at Berkeley, clicks away at a computer, retrieving the latest scan. The software flattens the cylinder to create one long, continuous picture of the grooves and their corresponding frequencies.

Because the 180-point, razor-thin laser is a tad more precise than a 19th-century stylus, the team can reproduce ranges of sounds the phonograph never could — high frequencies that left the thinnest of etchings. But that also means they pick up a considerable amount of noise.

Rabie — who joined Project IRENE as a research assistant during her freshman year and is now a Project IRENE analyst and processing specialist — is trained to sort out the static. Original phonographs couldn’t capture sounds above 5,000 hertz, for instance, so anything above that is noise she knows to digitally snip away. (The human ear can pick up frequencies from around 20 to 20,000 hertz.)

Still, every cylinder is unique — each one holds a piece of history, from tribal stories to dancing and healing songs to prayers. So Rabie reviews each scan, finding the range of frequencies specific to each cylinder. She sets parameters for the software, telling it what it to keep and what to ignore.

“Then, we listen again to make sure we didn’t lose anything,” Rabie says.

More complicated are the crevices left, most likely, by mold — artifacts of the cotton insulation that had surrounded the cylinders in cases for so long. In the scans, the supposed mold appears as a kind of spider web on the screen, with hairline cracks scattered over a section.

That’s where “blob cleaning,” as they call it, comes in. The grooves in a cylinder typically measure about 6 microns, or 6 millionths of a meter. (For reference, a strand of human hair is about 50 microns across.) Using new software by Cornell called Weaver — the “Photoshop for cylinders,” Battle says — the imaging team can inspect the damage and the surrounding region on a micron level and, rather than throw out bad points, add good ones, she says.

(Both the software and the project are named after the song “Goodnight, Irene,” by the Weavers — the team’s first-ever recording. IRENE also stands for Image, Reconstruct, Erase Noise, Etc.)

“Imagine you’re going along the groove, and the audio was supposed to be doing this, but now you have a dust particle,” Cornell says. “You can say, well, I’m good up to here, and I’m good here, and you can sort of draw a line across that section, if it’s small enough.

“If you miss several of these points, then you have to start getting fancier,” he continues. “You can look at the pattern before and after, and figure out what it would look like in between, and make it sort of match.”

In some cases, the damage is too tricky or too extensive to repair. In that case, the researchers leave the scans alone. The mission, Battle explains, is preservation.

“We can’t clean all of the scratching, and, in terms of balance, we always prioritize saving audio,” Rabie says. “It’s better to have a dirty cylinder than missing audio. The whole point is to have all the data that we can.”

Battle’s vision is that, as the technology evolves, they will be able to fine-tune their data further, opening ears to the recordings’ final mysteries.

Student talks about Project IRENE
Hands putting cylinders on scannerWax cylinders in Moffitt Library
Top to bottom: Clockwise from left: Youmna Rabie talks about her work on Project IRENE; Stephanie Battle sets up cylinders for scanning; and boxes that hold wax cylinders are lined up near the scanner. (Photos by Jami Smith for the UC Berkeley Library)

'A new surprise’

The team has scanned around 2,800 of the 3,000 cylinders in the Hearst Museum’s collection so far. The group scans three cylinders during the day and five at night. (Battle’s predecessor, Berkeley alum Olivia Dill, designed the system to run without an operator.)

“Some people might think that would get repetitive,” Battle says. “But every single one is a new surprise. Every single one is a new challenge.”

Rabie remembers a few in particular: the moving sounds of a man crying while singing, and a symphony of woodwind instruments spanning five cylinders.

“I was smiling the entire time,” Rabie says.

The project — funded by the National Science Foundation, the National Endowment for the Humanities, and the University of California — is set to wrap up in October. The recordings will be cataloged in the California Language Archive, an online database of the indigenous languages of the Americas. Depending on the sanctity of the material — many of the songs are considered sacred – some recordings will be shared with the public on the CLA website.

With help from the Hearst Museum and the Department of Linguistics, the recordings have already started making their way to the tribal and family members of those whose voices were captured long ago — many of whom cannot remember, and have no record of, their mother tongue.

“When I think about it, I just come in, click a few buttons, and type some code,” Rabie says. “But then I see it all working, I get to see the effects it has. … I’m proud of what I’ve put forward. It’s making a difference.”


Listen to two recordings below. The Rumsen voice belongs to Maria Viviena Soto, and the Salinan voice belongs to Pedro Encinales.