direkt zum Inhalt springen

direkt zum Hauptnavigationsmenü

Sie sind hier

TU Berlin

Inhalt des Dokuments

Source Separation from Stereo Musical Mixtures

Project Data
Project Manager
Prof. Dr.-Ing. Thomas Sikora
Funded by
TU Berlin
Project Period
02/2004 - 06/2007


Identifying and extracting the individual sound sources that are present in a mixture can be of extreme value for a wide range of semantic analysis applications (such as transcription, classification and denoising), as well as a powerful application by itself (unmixing). Motivated by the increasing potential of on-line music distribution services, we center our research on the separation of the instruments out of a musical mixture. We also emphasize on the stereo case, which is still the most common audio format. The goal of the project is to study the implications and requirements of such a separation, and to develop a system capable of identifying the number of instruments present in a mixture, to locate them spatially, and to resynthesize the separated sources. An important part of the research will be to study the combination of underdetermined Blind Source Separation (BSS) methods, such as sparsity- or ICA-related algorithms, with psychoacoustic-related methods, such as Computational Auditory Scene Analysis (CASA).

Prof. Dr.-Ing. Thomas Sikora

Dipl.-Ing. Juan José Burred

Zusatzinformationen / Extras