ISBN 978-3-8439-3858-7

978-3-8439-3858-7, Reihe Elektrotechnik

Matthias Brandt
Automatic Restoration of Audio Signals in Media Archives

149 Seiten, Dissertation Carl von Ossietzky Universität Oldenburg (2018), Hardcover, B5

Zusammenfassung / Abstract

A large number of historically relevant audio recordings are stored in media archives around the globe and represent an important part of mankind's cultural heritage. These recordings are stored on a variety of carriers, which usually underlie an aging process that results in the physical decay of the carrier material and ultimately the loss of the stored information. Therefore, large efforts have been made in recent years to digitize the inventory of media archives and prevent further deterioration. However, disturbances that have already been caused by the aging process remain in the digitized version of the recording. To reduce these disturbances, efficient audio restoration algorithms have been proposed, which typically require the manual adjustment of one or more algorithm parameters for each individual recording to achieve optimal restoration results. While manual operation is not a problem for a selected number of particularly valuable recordings, supervised restoration of complete archives is usually infeasible due to the sheer number of recordings.

The main topic of this thesis is automatic restoration of audio signals in media archives. More specifically, we propose algorithms that allow for an unsupervised restoration of a large number of audio recordings that show a great diversity with regard to the desired signal type, the disturbance type and the intensity of the disturbance. In doing so, we address three important disturbance types, i.e., impulsive disturbances, hum disturbances and broadband noise. Impulsive disturbances frequently occur with grooved recording media, e.g., wax cylinders, shellac and vinyl discs, and are caused by dirt and mechanical deformations of the media. Hum disturbances are often caused by power line interference with the audio signal during the recording process. Broadband noise is mainly caused by restrictions of the recording medium, e.g., the size of the magnetic particles for tape media. A key element in the design of the proposed algorithms is the desire to keep the degradation of the desired signal as low as possible while achieving a substantial improvement of the audio quality.