The history of recorded music is worth preserving for future generations. However, the earliest examples of the recording arts are difficult, if not impossible, to hear anymore. Many wax cylinders and shellac disks are crumbling in archives, unable to be played because any physical contact with a stylus would cause them irreparable damage. Even those that can be played often suffer from surface noise and scratches that cause clicks and pops. Additionally, many are broken, making even the most careful stylus-based playback impossible.
FIG. 1: This section of a 78 rpm shellac disk measures 1.39 5 1.07 mm; the groove is 160 µm wide. The lateral displacements along the length of the groove represent the audio signal. Audible effects created by dust and scratches can be removed using data processing after the surface is scanned.
An ingenious solution to these problems was recently discovered by Carl Haber and Vitaliy Fadeyev, physicists at the Lawrence Berkeley National Laboratory (www-cdf.lbl.gov/~av). They realized that the optical scanning techniques used in precision detector arrays and the statistical methods used to analyze the tracks of particles emerging from collision experiments could be adapted to extract audio information from old disks and cylinders.
From the earliest recordings of Thomas Edison to the monaural 78 rpm disks of the 1950s, audio signals were mechanically encoded directly onto the medium. On a disk, a spiral groove was cut into the surface, and audio information was represented by lateral displacements in the groove from its nominal path. In other words, the groove wiggles from side to side on what would otherwise be a simple spiral path. On a cylinder, the groove forms a helix that wraps around the cylinder from one end to the other, and the displacements that represent audio information are vertical, forming small hills and valleys along the groove, rather than lateral wiggles.
To play the audio, a stylus rides in the groove while the disk or cylinder turns at a constant rate. As the stylus follows the wiggles, its velocity changes in a pattern that is analogous to the original audio waveform, which is reconstructed by the playback device.
After hearing about the problems of preserving mechanical audio recordings, Haber and Fadeyev realized they could optically scan the surface of a disk or cylinder and process the resulting data much as they do to reconstruct the paths of particles created in beam collisions. Avoiding all physical contact with the medium, they could remove the effects of scratches, dust, wear, and surface noise, using sophisticated data processing.
Their initial experiments were conducted on 78 rpm shellac disks from the 1950s. Using a computer-controlled optical scanner with high resolution, they imaged small sections of the disk surface (see Fig. 1). Each image generated about 1 MB of data; the audio information on an entire 10-inch, 78 rpm disk amounted to 100 to 1,000 GB of data before processing. The computer stitched together the images and analyzed the groove wiggles, deriving the stylus velocity while removing spurious data and reconstructing the audio signal in digital form as 44.1 kHz, 16-bit WAV files.
These experiments were a big success as a proof of concept, but the equipment was not optimized for the task; the process was slow, taking 40 minutes to yield one second of audio. However, Haber and Fadeyev believe that a dedicated system could dramatically reduce the time frame to somewhere between 5 and 15 minutes for a 10-inch 78 rpm disk. In addition, their 2-D imaging cannot be used to scan cylinders; 3-D scanning is required to measure the hills and valleys of the groove's displacement. Haber and Fadeyev have already started experimenting with such a system, and the initial results are promising. Clearly, this work could help rescue our musical heritage from oblivion, allowing future generations to learn from the past as they create the music of tomorrow.