Reading Record Grooves

I found a preservation technique that photographs a record's groove and reconstructs its audio without ever dropping a needle. A camera captures the microscopic contours of the recording, and software reconstructs the motion a stylus would have made.

A record stores sound as physical variations along a groove. On a lateral-cut disc, the groove meanders from side to side; on some older recordings and cylinders, its depth changes instead. A record player reads those variations with a needle and converts its movement into an electrical signal.¹

There are other examples of this kind of physical encoding. A player-piano roll uses punched holes to determine which notes an instrument plays and when. It can reproduce a musical performance, but it contains instructions rather than a recording of the sound itself.²

Film used for motion pictures takes a different approach: its soundtrack sits beside the picture frames as a narrow strip that varies in width or darkness. Light passing through that strip reaches a sensor, producing an electrical signal that can be played as sound.³⁴ On a compact disc, a laser reads microscopic pits and lands, and the player decodes the resulting digital information into audio samples.⁵

At the Library of Congress, a preservation system called IRENE creates high-resolution images of recorded grooves instead of running a needle through them. Flat imaging captures the sideways variations on a disc, while three-dimensional scans measure the changing depths of vertical-cut recordings and cylinders.⁶⁷

The camera supplies a measurement of the surface, rather than a finished audio file. The audio is reconstructed using software, which has to identify the groove, follow it through scratches or dust, and estimate the movement a needle would have made. It also needs a playback speed: the same groove produces a different sound depending on how quickly it passes beneath the needle. The process resembles remote sensing, where an instrument gathers information without touching what it is measuring.

Image-based recovery can even make recordings audible when they were created without any playback mechanism. In 1857, Edouard-Leon Scott de Martinville introduced the phonautograph, a device that traced airborne vibrations onto a surface so they could be studied visually. In 2008, researchers digitally reconstructed one of his 1860 phonautograms and made it audible 148 years later.⁸⁹

A four-second sound can be encoded as a groove in an image, then recovered by software that traces its shape. The example below compares two versions of the same groove: one clean, and another marked with scratches and dust.

Scratches, dust and other forms of damage make decoding more difficult as they introduce loss or noise. The following is a closer look at what happens in the presence of these disturbances.

The three clips below make the comparison easier to follow: the original sound, the version recovered from its clean groove, and the version recovered after the groove was damaged.

For this signal, the clean reconstruction reached a 0.990 correlation with the original. With scratches and dust, the correlation dropped to 0.979 and the signal-to-error ratio fell from 17.2 to 13.7 decibels.

A camera cannot replace parts of a groove that no longer exist, but it can preserve what remains without wearing it down. A recording and its playback equipment no longer have to survive together: if enough of the physical pattern remains, software can reconstruct its sound, even when the original format was never designed to be played.¹⁰