VisualAudio


VisualAudio is a project that retrieves sound from the picture of a record. It originated from a partnership between the Swiss National Sound Archives and the School of Engineering and Architecture of Fribourg

Introduction

Disks were the only mean of preserving sound before the introduction of magnetic tapes. Until the advent of the vinyl in the 1950s, the records were made of shellac or wax. The organic composition of these materials enabled them to degrade over time and also made them prone to attack by fungi.
As a result, many records, including unique original radio productions are in a state of deterioration which precludes play by traditional mechanical means. Hence the interest in a non-contact approach.

History

The idea of this recovery of the sound of old records through optical scanning started in the summer of 1999 in Lugano, among the technical manager of the Swiss National Sound Archives Stefano S. Cavaglieri, the former director of M & C Management and Communications SA Pierre Hemmer, and the Director of the Swiss National Sound Archives Pio Pellizzari.
The Fribourg school of engineering and architecture was the main partner, at first studying its feasibility and then starting a project that progressed over years.

Principle

During normal playback of a phonograph record, sound is obtained by a stylus following the groove.
The radial displacement of the groove is observable through a microscope which means that sound information is visible.
If a high-resolution analog picture of each side of the record is taken and the information in the film is then digitised using a circular scanner, various algorithms can process the image in order to extract and reconstruct the sound.

Method

Photography

A central part of the process is photo shooting. It is performed at the beginning of the process on a properly cleaned record, in order to archive it as a film.
The photographic film has a high resolution of 600 lines per mm. This resolution is sufficient to accurately follow the groove displacement.

Scanning process

Once the record content is stored on a photographic film, the next step is to recover the original sound. The University of Applied Sciences of Fribourg built a scanner prototype to do this.
The current version of the scanner is made of a glass rotating plate, on which is placed the film. The digitisation of the image is done by a linear CCD camera of 2048 pixels wide, which takes pictures at regular intervals, with frequencies ranging from 25,000 to 200,000 lines per rotation. The combination of the camera with the rotating film delivers a rotary scan of the record in the form of a rectangular picture of a ring. A second radial movement provides the next ring.

Image processing

Once digitized, the images are processed to analyze and determine the positions and displacement of the groove. The first step is to correct the imperfections of the captured images. Many disturbances can come from various stages of the acquisition process: the record itself, the photography, or the scanning.
Then, the groove position is estimated using edge detection algorithms. Once the edges are detected, corrections requiring more complex knowledge about the structure of the image are carried out. Some examples of corrections:
The final step is converting groove displacement into an audible signal. This signal is processed by band-pass filters in order to obtain only the bandwidth of the original recording. Some frequency equalizations are implemented.
This project aims to retrieve and archive the sound as close as possible to the original one. By default, audio restoration is not applied.

Broken records

The ultimate goal of this project is getting sound out of an otherwise forever lost record.
Many records from the 1940s are cracked and definitely unplayable. The result is an interesting jigsaw puzzle.
As the cracks are due to the shrinkage of the lacquer, there is no material loss in most of the cases. To solve this problem, in November 2006, the Swiss National Sound Archives started a project funded by the Gebert Rüf Foundation. The results so far are encouraging. The algorithm basically uses signal features to determine whether two groove parts are contiguous or not.
The project is still in a test validation phase, but some sound is already available.

System quality

Reaching the same quality as good as an original record replayed on a modern turntable is probably unrealistic. Originally around 20 dB in the early prototype, the signal to noise ratio of modern system is situated around 19 dB for a good 78 rpm record.

Advantages and Disadvantages

With its intermediate photographic stage, this solution solves several significant challenges found in archiving systems.
Disadvantages of the system:
Among the unique audio files recovered with such techniques, the speech of Italian politician and poet Aldo Spallicci.