certain angle of incidence, describes the sound transmission from a free
field source to a point in the ear canal of a human subject. Knowledge
of human HRTFs is essential in the design and evaluation of artificial
heads and in computer synthesis of binaural signals.
Because of anatomical differnces between people, there are individual
differences in human HRTFs. It cannot be expected that any
non-individualised HRTF will produce good results for every person.
Binaural signals processed using one's own HRTF give the most realistic
listening experience, but extensive and expensive acoustic measurements
are involved in obtaining them. Numerical methods have proved
to be a
possible alternative to acoustic measurements. However, they
are
normally computation intensive.
In this project we seek to create acoustically authentic boundary
element models of the head and shoulders, derived from 3D images of
human subjects. A programme of progressive refinement and validation
of
the technique is underway. We estimate that reliable simulation
of
acoustic wave propagation around the head will be possible up to at
least 11.5 kHz. The model will be a powerful tool in acoustical
studies
of human hearing, especially providing new insights into the link
between the human form and sound localisation mechanisms. We
propose to
use the model to synthesise families of HRTFs for several normal
subjects. By means of perceptual measures and formal listening
tests
these image-derived personalised HRTFs will be validated against
personalised HRTFs measured using the conventional acoustic method.
A
successful outcome to this work will contribute to the development
of a
simple image-based process for creating accurately spatialised sound
tailored to an individual. The ultimate convenience of this approach
is
expected to lead to greatly expanded applications of 3D sound in a
wide
range of consumer electronic products. Possible applications
include
sound spatialisation products for multimedia, telepresence, virtual
reality, the internet, games software and edutainment software.
The
work will also be of relevance to sectors of the recording and broadcast
industries concerned with improving the fidelity of reproduced sound.
In next-generation hearing aids, spatialised sound will improve a user's
ability to select and reject sounds in a noisy environment.
Initially, head shape will be obtained by using 3D cameras and laser
scanners. In preliminary work we have made plaster-of-Paris casts
of
our ears and have used a CT scanner to obtain detailed pinna surface
descriptions.