Optical Metrology Reconstructs Audio Recordings
Daniel S. Burgess
Under an interagency agreement, the Library of Congress in Washington and Lawrence Berkeley National Laboratory in Berkeley, Calif., will investigate the application of optical metrology to the preservation of audio recordings. The researchers on the project hope that the approach, which was inspired by methods that measure traces in particle detectors, will enable the capture of recordings on deteriorating wax cylinders and 78-rpm discs in library collections around the world and will ease the digitization of these materials.
The lab's Vitaliy Fadeyev and Carl Haber have demonstrated that the surface map generated by a metrology system of a mechanical recording can be used to reconstruct the audio information. Image processing tools identify and merge the frames containing features of interest to generate a data set describing the 2-D or 3-D shape of the groove encoding the information (Figure 1). A mathematical algorithm applied to these metrology data yields an approximation of the playback signal that a stylus applied to the recording would produce, which may be saved as a digital audio file.
Figure 1. Researchers are investigating the use of optical metrology techniques for the preservation of audio recordings. The data from two-dimensional digital images of the surface of a record can be used with an algorithm simulating the behavior of a stylus on the disc to produce a digital audio file of the recording. Image processing can compensate for some sources of noise, such as small scratches and dust.
Libraries worldwide are investigating strategies to ensure that recordings from the dawn of the phonographic age will be accessible to future generations. "The preservation of the audiovisual record is the preservation of the language of the 20th century," said Mark Roosa, director for preservation at the Library of Congress.
The Library of Congress' recorded sound collection comprises approximately 2.5 million examples of vocal and instrumental music, radio broadcasts, speeches and interviews. Featuring media ranging from wax cylinders to minidiscs, the collection adds an average of 100,000 recordings per year, said Samuel Brylawski, head of the recorded sound section of the library's motion picture, broadcasting and recorded sound division.
Some of these media are relatively stable, although all will degrade over time. The earliest recordings, which were produced in the late 19th century and stored on so-called brown wax cylinders, suffer from "inherent vice," Roosa said. The material chemically deteriorates, crystallizes to fill in the grooves that captured the sound information and can grow mold. Early discs also suffer from degradation problems that stem from their being constructed of a composite medium, a plastic or shellac film over a filler. Because of the different material properties of the surface layer and the filler, changes in temperature and humidity lead to delamination and cracking of the recorded surface. Moreover, methods of playback become obsolete, an issue especially true today for cylinders, he said.
Preservationists thus desire a relatively nondestructive means to acquire the recorded audio information for storage in a form that is very stable – to survive the ravages of time – and that may be migrated forward as playback technologies change. Such an approach also must be scalable to address the challenge of applying it to massive holdings in the world's collections. Consequently, the library was very interested when Fadeyev and Haber approached it with the idea of using optical metrology systems to digitize recorded materials.
Fadeyev and Haber initially dem.onstrated that the Avant Zip 400 Smart Scope from Optical Gaging Products Inc. of Rochester, N.Y., could produce 2-D maps of 78-rpm records sufficient to reconstruct audio recordings. The general-purpose metrology system, which the lab is using to build particle detectors for the CERN Large Hadron Collider, comprises a precision servo-driven X-Y table and a 1/2-in., 768 × 494 color CCD camera with an autocalibrating zoom lens. For this application, built-in software tools performed edge-finding on the record groove and created an image features file.
In experiments with samples obtained from the used-records bin at a local music store, they processed the image features data offline, re-sampled the resulting audio data at 44.1 kHz and saved the data in .WAV format. Comparisons of optically reconstructed, mechanically played and reissued digitally remastered versions of one of the samples, the Weavers' "Goodnight Irene," revealed that even this nonoptimized setup yielded a faithful reproduction (Figure 2).
Figure 2. A comparison of optically reconstructed, mechanically played and reissued digitally remastered versions of one recording revealed that even the nonoptimized 2-D setup yielded a faithful reproduction.
Since then, the researchers have used a CHR 150 color-coded confocal imaging system from Stil SA of Aix-en-Provence, France, to optically reconstruct the wartime ballad "Just Before the Battle, Mother," performed by Will Oakland and Chorus, as recorded on a 1920s-era celluloid cylinder. (Called Blue Amberol, the medium replaced brown and black wax and represented the last hurrah of the phonographic cylinder.) Color-coded confocal imaging employs a polychromatic light source and optics with exaggerated chromatic aberration. Different wavelengths from the source focus at different depths on the sample, effectively creating a system that assembles a series of wavelength-dependent 2-D surface profiles to generate a 3-D map of the surface (Figure 3).
Figure 3. Using color-coded confocal imaging, the researchers generated three-dimensional maps of the surface of a 1920s-era celluloid phonographic cylinder from which they reconstructed an audio recording.
The scan was performed at Tai-Caan Technologies Ltd. in Southampton, UK, in collaboration with Christian Maul of TaiCaan; John W. McBride, also of the company and of the University of Southampton; and Mitch Golden. In this case, Fadeyev and Haber processed the topographic data offline and performed resampling at 22.05 kHz. A comparison of the optically reconstructed version, saved in .WAV format, with one captured mechanically from a commercial reissue of the recording revealed no qualitative differences.
Under the agreement, the researchers hope to develop dedicated systems with much higher sampling rates. Although the Avant Zip 400 Smart Scope, for example, demonstrated the feasibility of the concept, Fadeyev explained that its sampling rate in this application was limited to approximately 66 kHz. In contrast, a specialized machine could achieve rates six times higher, potentially enabling the user to scan a 10-in. 78-rpm record in five to 15 minutes. Three-dimensional techniques are slower, and scanning a cylinder may take several hours.
Other avenues of study, Haber said, will include determining if the technique can recover recordings from moldy or damaged cylinders. One benefit of digitization is that it enables preservationists to compensate for the effects of deterioration to recover the original sound or even to reassemble the digitized pieces of a broken recording. And the technique offers users the ability to perform noise reduction and scratch repair as part of image processing.
"The more and more we talk about the possibilities, the more and more we're excited," Roosa said.
MORE FROM PHOTONICS MEDIA