FEATURE
Voices From the Crypt: IRENE Preserves ‘Groovy’ Audio From the Past
by Jan Zastrow
What Is IRENE?
In a nutshell, the IRENE (image, reconstruct, erase noise, etc.) process uses a “touchless” optical-scanning technology to create a high-resolution digital map of a grooved record or wax cylinder and, by means of software, that calculates the motion of a stylus moving through the grooves, replicates the audio content, and produces a digital sound file. |
| Now, without even having to touch the audio relic—and risk the pressure of a stylus—an image of the groove pattern can be made and the sound extracted using computer software. |
Imagine hearing sounds recorded more than 130 years ago that we thought were lost forever: the voice of Alexander Graham Bell testing his equipment by reciting numbers in 1885,1 the speech of King David Kalakaua—Hawaii’s “Merrie Monarch”—4 days before his death in 1891,2 or a member of the Native American Hupa tribe telling stories in his native tongue in 1901.3 Impossible, you think … but you would be wrong.
Old audio recordings often present us with obstacles, including the inability to play them back or to reconstruct sound from broken discs and cylinders. Experiments with recording sound have been going on since the 1860s, but it wasn’t until 2002 that two physicists at the Lawrence Berkeley National Lab (LBNL) in California—Vitaliy Fadeyev and Carl Haber—tried scanning records with an imaging system they normally used for physics experiments. They discovered that this method (optical metrology) could be used to measure the grooves in vinyl discs, wax cylinders, dictation discs, and a wide variety of other historical media.
Why is this so thrilling to archivists, oral historians, ethnomusicologists, Native communities, and others who are interested in antique sounds? Anything you can embed noise in can get scratched, bent, broken, torn, warped, or melted; many old audio recordings are so fragile that playing them would destroy them. Customarily, a stylus would trace the motion in the groove of a disc or cylinder while it’s rotating, and then the sound would be amplified either acoustically using a horn (think Gramophone) or electronically transmitted to be amplified and sent to loudspeakers. Now, without even having to touch the audio relic—and risk the pressure of a stylus—an image of the groove pattern can be made and the sound extracted using computer software. Broken discs and cylinder records can be temporarily reassembled to capture an image and then processed into an audio file.
Deep Dive
For the scientific among us, here’s the long version. Peter Alyea, of the Library of Congress, writes, “The IRENE System is equipped with two different imaging methods to profile the groove motion. The two-dimensional system is a specialized microscope that focuses a bright white light directly down onto the surface of the audio media. In order to understand what sort of picture results from this lighting, one needs to understand that the groove is made up of the land (the highest part of the disc surface), sidewalls (the angled parts where the audio information is encoded) and bottom. When you shine a light down on the media, the areas that are perpendicular to the beam reflect the most light back to the source, due to the directionality of light. The motion of the groove is then revealed in the contrasting intensities of the land, sidewalls and bottom.
 |
 |
“The IRENE System uses a unique sort of two-dimensional camera that images a single pixelwide line across the grooves. It takes many pictures in succession as the media is in motion. A typical scan will result in 80,000 images per rotation, which covers around 10 to 12 groove rotations. The camera is moved to take multiple rotations until the entire surface of the grooved media has been imaged. Because the camera images only a single line at a time from a fixed position, the circular curve of the groove on a disc is “unwound” into a series of straight ribbons (see Figure 1). The contrast between the light areas of the land and bottom and the dark areas of the sidewalls contain the motion information needed to produce the sound.
“By following the edges of the contrasting areas of the image, the positional information is extracted by the PC workstation. The image strips are then analyzed by custom software that identifies the motion of the groove and converts that to sound. The motion in a groove is quite small so the images taken are on the micron and sub-micron level; by way of comparison, a human hair is approximately 50 microns.
“To take three-dimensional images for a cylinder record, a line of white light shines on the surface through a lens (based on the principle of chromatic aberration) that focuses different frequencies of light at different, but predictable, distances. This creates a 3-dimensional profile that is then analyzed to extract the motion of the groove and calculate the sound. Three-dimensional imaging provides a redundancy of data which can result in less ‘noise’ in the audio. The downside is that imaging in three dimensions with the currently available hardware is quite a bit slower than the two-dimensional method.” (See Figure 2.) |
This system—developed in conjunction with the Library of Congress’ (LC) Peter Alyea and others—is called IRENE (image, reconstruct, erase noise, etc). Why IRENE? The name was chosen because the first successful disc scanned by Fadeyev and Haber was “Goodnight, Irene,” recorded by The Weavers in 1950.
The Technology Behind IRENE
I caught up with Alyea at the LC recently and asked him to explain the technology behind IRENE. His short answer was, “The IRENE System takes digital images of the media while in motion, and then processes those images to produce the sound. That means that even media which is badly degrading or broken can be repaired, reassembled or otherwise manipulated in the image domain before processing to an audio file.” Alyea went into some very scientific explanations describing 2D and 3D imaging. Luckily, there’s an article from The New Yorker that clarifies this: “Analog recordings—that is, cylinders and records—have two kinds of grooves, lateral and vertical. Records have lateral grooves and cylinders have vertical ones. Lateral grooves move side to side, like a river through a plain; vertical grooves rise and fall like hills. … Three-dimensional scanning registers more of the groove, and is more precise: two dimensions is a map; three dimensions is a topographical map.”4
Components of the IRENE System
To perform all of this work, clearly, one needs very specific hardware and software. What does it take to put together an IRENE system? At a cost of up to $200,000 apiece for the Cadillac version, there are only six in the world: one at LBNL; two at LC facilities; one at the Northeast Document Conservation Center (NEDCC) in Andover, Mass.; one in India at the Roja Muthiah Research Library; and the newest at the University of California–Berkeley’s (UC–Berkeley) library.
 The specialized software used in the IRENE system—the “secret sauce” programmed by the whizzes at the LBNL and the LC—includes capture software that controls motion stages as well as image capture and image-to-sound processing software that analyzes images to calculate the motion of the recording and extract sound (see Figure 3).
Necessary hardware comprises three main components: The computer-controlled motors that move the media and the optics, the laser profile sensors used for focus control, and the 2D and 3D imaging cameras. They all work together to gather image data that is fed to a PC workstation to extract the sound.
Outsourcing
If your historical audio collections aren’t quite so vast as to require a dedicated system of your own, the NEDCC offers a variety of a la carte services. In addition to IRENE digitization of grooved media into WAV and TIFF files, it offers an intriguing “discovery” service—30-second clips of audio—in order to assess content for, say, unlabeled media when prioritizing preservation needs or writing a grant application.5 (It’s worth visiting the NEDCC website if for no other reason than the great explanation, in pictures, of how the IRENE system works.6)
Speaking of grants, the Council on Library and Information Resources (CLIR) is presently administering a grant (Recordings at Risk), funded by the Andrew W. Mellon Foundation, through September 2018. Its aim is “to support the preservation of rare and unique audio and audiovisual content of high scholarly value through digital reformatting.”7 Awards range from $10,000 to $50,000 and cover costs of preservation reformatting by qualified external service providers, such as NEDCC. If your institution has significant at-risk scholarly audio collections, there’s still one more round of applications, so it’s not too late.
The Future of IRENE
But back to groovy audio media and optical scanning: How is this being used now, and what might we expect in the future? The latest application for optical scanning is Documenting Native Languages at UC–Berkeley, a 3-year project that is scanning 3,000 wax cylinders in the Hearst Museum of Anthropology collections.8 This will provide access to early 20th-century ethnographic field recordings that record native Californians singing and speaking in native languages.
The 2.0 version of the IRENE software provides a new, extensible architecture that allows others to contribute and add to it. Currently, students are writing plug-ins to analyze the geometries of different media, line up grooves of broken media, measure the centricity of the grooves on a disc, restore the image to produce improved sound, and more. This new software may eventually become open source, so one day, the media could be scanned at a service bureau that owns the IRENE equipment. Then the institution would get the image data to manipulate, play back, share, store, or other up-to-now unimaginable purposes. Reproduction of discs and cylinders via 3D printing is also being explored, although it can’t yet print at a high enough resolution for the tiny motions on the grooves.
Preserving the data on the physical object allows it to be repurposed for other creative outputs. Imagine those 2D and 3D groove maps being set to colored lights, recorded poetry, or … well, fill in the blank. Twenty years ago, if someone said we would be able to scan a record, who would’ve believed it? If this can happen, the sky’s the limit.’
Many thanks to Peter Alyea, digital conversion specialist in the Library of Congress’ Preservation Reformatting Division, for the advice and technical expertise he lent during the preparation of this article.
|
