Summary: ON­LINE BLIND SEPARATION OF MOVING SOUND SOURCES
J˜ orn Anem˜ uller and Tino Gramss (y)
Dept. of Physics, Graduate School in Psychoacoustics
Carl­von­Ossietzky­University, 26111 Oldenburg, Germany
ane@uni­oldenburg.de
ABSTRACT
In this paper, we propose a method for the on­line blind sep­
aration of sound sources in the case where the mixing filters
have a Æ­shaped impulse response. Our algorithm works en­
tirely in the frequency domain and exhibits fast convergence
due to cross­frequency couplings.
Specific problems related to on­line separation of run­
ning speech are discussed. The algorithm performs success­
ful separation of digitally mixed speech signals and of sig­
nals from a moving and a standing speaker recorded in an
anechoic chamber.
1. SOURCE SEPARATION
In the case of an instantaneous linear superposition
m i (t) =
X

Source: Anemüller, Jörn - Medical Physics Section, Carl von Ossietzky Universität Oldenburg

Collections: Biology and Medicine