Distributed Digital Music Archives and Libraries (DDMAL)


The research intends to focus on access to and long-term preservation of analogue sound recordings. There are several compelling reasons for which the digitization of analogue sound recordings must be initiated. Although many analogue recordings have a long shelf life, they do degrade upon playback. Some media, such as magnetic tape, are deteriorating rapidly. Many rare and important recordings are being discarded, and the appropriate playback equipment is becoming scarce. For example, there is only one currently manufactured system that can play wax cylinders. Much research is needed to ensure the proper digitization and preservation of these very important cultural heritage materials.

Because of the sheer volume of analogue sound recordings in existence, there is an urgent need to begin contemplating the proper method for digitization. The British Library has catalogued over 2.5 million sound recordings (even with a conservative estimate of digitization time at 30 minutes per recording, the total digitization time required would be 1.25 million hours, 52,000 days, or 142 years). The Library of Congress has 2.5 million recordings of various kinds that do not overlap significantly with those of the British Library. Music libraries worldwide hold tens of thousands of recordings (e.g., the music library at the University of Illinois at Urbana Champaign holds 115,000 recordings). As of July 2003, OCLC WorldCat contained 1.7 million catalogue entries primarily for commercial sound recordings.

Two different methods will be considered for the digitization of phonograph records: one mechanical and the other optical. In the traditional mechanical means of sound reproduction using a stylus, one of the major challenges consists in choosing the appropriate analogue-to-digital (A/D) conversion system (turntable, tonearm, cartridge, preamp, interconnect cables, A/D converter, etc.) to achieve archival-quality results. The wide range of opinions on this subject, especially among the “golden ears,” constitutes a major obstacle for many digitization projects. Most audio digitization projects undertaken thus far have concentrated on access rather than on preservation. If recordings are digitized for the purpose of long-term preservation, extreme care must be taken so that the process need not be repeated in the near future. The sound quality of various combinations of components in the digitization process will be evaluated through psychological experiments designed to determine, for example, whether there are perceptual differences or preferences.

A radical alternative means of digitization is to optically scan the phonographs for preservation and subsequently convert the scanned image to audio for access. There are tremendous advantages to preserving phonograph recordings as 2D or 3D images. This would make it possible to experiment freely with methods of converting to audio—with a view to determining the best method—without touching the recordings themselves.

Early this year, two physicists, Vitaliy Fadeyev and Carl Haber, from Lawrence Berkeley National Laboratory (LBNL) in California, in a ground-breaking experiment, successfully produced high-quality sounds by optically scanning 78rpm records with video microscopes and then converting the images to audio (Fadeyev and Haber 2003). Previously, scientists used laser beams to track record grooves. This is the first time that a 2D-scanned image was used to produce quality audio. The incredible achievement was possible because of recent developments in high-resolution video inspection systems used in medicine and metrology. These devices possess sub-micron resolution, which is necessary for resolving the details found in record grooves. This technology has tremendous potential for audio preservation.

The initial LBNL experiment was performed on a few 78rpm recordings. Further experiments will be conducted with the infrastructure to improve and extend this technique for digitizing wax cylinders and stereo LPs. Since, in most 78rpm records, the grooves are cut laterally, parallel to the surface of the record, a 2D scan can be used to extract audio information. The acquisition of the video microscope for the infrastructure will enable the applicant to find a method for obtaining audio from 3D images, since wax cylinders store information vertically, perpendicular to the record surface, and stereo LPs use both lateral and vertical indentation to store two channels of audio. Another important research goal using the microscope to optically scan the discs is to reduce the amount of time required to perform the scan. Fadeyev and Haber report that their scan took 50 minutes for 1 second of audio (3000 times the real time). They do note, however, that their scan was not optimized for time. Despite these challenges, there are several significant advantages to this approach that warrant further investigation. Optical scanning is possible even if the disc is mechanically unplayable because it is too fragile, too valuable, or broken. The effects of wear caused by stylus may be overcome by scanning the region away from the inner surface of the groove where the stylus normally tracks. Traditional sound restoration uses filters in the time or frequency domain. With imaging, the quality of sound can be improved by removing artifacts in the spatial domain where the noise originates. In general, imaging captures more data than laser beams or mechanical stylus, thus allowing more flexibility during the restoration process. The image files can be stored as preservation copies and used as the basis for improved restoration techniques in the future. The plan is to acquire a newer version of the microscope that was used at LBNL, OGP Quest 450, which has improved resolution and speed. It will be equipped with a rotation indexer for wax cylinder scanning and with a DRS laser to facilitate the 3D scanning.

The result of the image-to-audio conversion will be quantitatively and qualitatively compared with the best results obtained using mechanical stylus. Furthermore, much effort will be invested in developing software for audio restoration, which attempts to remove pops, clicks, and other noise from older recordings. Other challenges include determining the equalization curve used during the recording of 78 rpm discs, since it was not until later that recording industry agreed on using the same equalization curve, called RIAA, when producing the records. The speed of recording was also not standardized thus the problem is correctly determining the playback speed. It is hoped that these can be solved using digital signal processing techniques.

This project will proceed in parallel with the digitization of McGill music library’s 78rpm Jazz recording collection (funded by a three-year FQRSC research grant) and digitization of a unique collection of Handel LP recordings (funded by McGill’s Richard M. Tomlinson Digital Library Innovation Awards). See MAPP.



Fadeyev, V., and C. Haber. 2003. Reconstruction of mechanically recorded sound by image processing. LBNL Report 51983.

Created: 2004.04.11 Modified: Ichiro Fujinaga
McGill Crest