New Semiautomatioc Denoising Algorithms for Historic Music Recordings

Tutor: Stefan Goetze
Type of Thesis: Diplomarbeit (Dipl.)
date of end: 12/2008
Student: Matthias Brandt
Status: finished
ANT-shelfmark: D-12/08-1

This thesis presents a method for reducing additive noise in historic audio recordings. The focus is on the restauration of music recordings. It turned out that most of the established denoising procedures, that aim at processing noisy speech signals, are inapplicable for operating on music signals without further modification. Thus, enhancements and modifications of the common methods are introduced, and solutions for emerging problems are presented.

Furthermore, the denoising process was automated to a certain extent -- the required user interaction has been highly reduced in contrast to existing methods.

Achieving the highest possible sound quality was of great importance in the development of the system. To analyse preferred parameter setups and, finally, the efficiency of the resulting denoising method, extensive listening tests were conducted, in which test persons evaluate the quality of the output signal.


Last change on 15.12.2008 by S. Goetze
AIT ieee GOC tzi ith Fachbereich 1
© Department of Communications Engineering - University of BremenImprint / Contact