|
 |
Uni
Faculties
Fk. V
Physics
Medical Physics Demos
Audio Demonstration of physiologically inspired signal
processing: Use of Noise-Robust Sound Localization for the
Separation of Voices
This demonstration shows the application of an algorithm which
was inspired by physiological findings in the barn owl. The
algorithm is described in chapter 3 of the PhD thesis
"Localization and separation of concurrent talkers based on
principles of auditory scene analysis and multi-dimensional
statistical methods". The statistical sound localization
algorithm, on which it is based, is described extensively in
chapter 2 of the same work, and has been published in Nix,
J. and Hohmann, V., (2006), Sound source localization
in real sound fields based on empirical statistics of interaural
parameters, J. Acoust. Soc. Am. 119 (1).
In addition to the a priori statistic of interaural
parameters in noise, the algorithm uses the values of the
interaural transfer function (ITF) to separate the voices. The
sound source localization algorithm has a convergence time of
about 0.2 s; For the estimation of the filtering coefficients,
a larger time constant is used to reduce artefacts in form of noise
bursts.
Both parts of the algorithm are easily extensible to situations
with more microphones and more talkers; for the separation of sources,
not more sources than microphones should be present. For an efficient sound
localization, it is beneficial that the pairwise
inter-microphone transfer functions are somewhat asymmetrical in
space.
(more publications about sound localization)
|
print