Problem
Sound source detection is based on certain sound properties, such
as pitch and spectral shape, which must be estimated from a
SIgnal. We try to mimic human auditory perception, as we aim to
achieve human performance. So far the outer ear, middle ear,
cochlea and part of the hair cells and auditory nerve have been
modeled. However, several of the neural processing steps have been
modeled sketchily at best.
Possible solution
Several toolboxes are available containing models of several parts
of the auditory neural pathway. Most of these models are complex
and would require too much computational effort to be implemented
in real-time. However, a simplified version still containing the
essential characteristics may be possible. Also, the effect of
modeling the neural pathway in more detail may be that new
detection criteria for several sound categories can be found.
Proposal
We would like to try to compare the output of several of the
models for various parts of the auditory neural pathway. A
detailed study of their outcome on a database of natural sounds
may provide more insight into human detection methods and
criteria. If useful, a simplified version of these models may be
included in future applications.
Contact
Ronald van Elburg ()
postal address
Auditory Cognition Group
Department of Artificial Intelligence
University of Groningen
P. O. Box 407
9700 AK Groningen
The Netherlands
visiting address
Bernoulliborg
Nijenborgh 9
9747 AG Groningen
