We are able to locate sound by recognising three main auditory cues:

Interaural Time Difference (ITD) - sounds from one side of our heads will arrive at one ear before the other, so there is a time delay between ears

Interaural Intensity Difference (ITD) - the ear opposite the sound source will not recieve as much sound, so there is a volume (intensity) difference between ears

Spectral changes - sounds change as they travel around the head, and particularly as they travel past the pinnae (the fleshy external part of the ear), leading to changes in the sound spectrum as recieved by the ear.

These three cues can be encapsulated in a Head Related Transfer Function (HRTF). With full knowledge of the HRTF, we can process sounds such that they appear to come from wherever we wish, but unfortunately the HRTF varies from person to person due to physiological differences between heads.

If we measure an individual's HRTF, we can "personalise" the reproduction system for that individual. Current HRTF measurement methods rely on acoustic means, which are slow and costly. A better knowledge of how head-shape influences the HRTF might allow for methods based on physiological measurements to be used instead. The rest of this page details the processes being developed by the research group to acquire and modify head models, and to calculate what effects the modifications have.

Acquisition
The head of the KEMAR mannequin is used since it is a standard reference. The head is digitised using a Fastscan laser scanner, which results in a point cloud which is then meshed. The ears are scanned seperately using Magnetic Resonance Imaging (MRI) or Computerised Tomography (CT), courtesy of York Neuroscience Centre and York District Hosptial respectively. The meshes are then cleaned up, and the ears attached to the head, currently using Autodesk 3ds max.

Parameterisation
In order to allow easier mathematical manipulation of the shape, the head mesh is then parameterised using an Elliptic Fourier Transform. More information about the derivation and use of this transform can be found in the references on the research page.

Peturbation
The parameterised representation of the head is then peturbed to change its shape very slightly. Multiple, random, peturbations are performed and stored

Simulation
The HRTF of each peturbed head is then calculated by reconstructing the mesh from the Fourier coefficients and using PAFEC acoustic simulation software to run a Boundary Element Method analysis. The computational horsepower is provided by the White Rose grid computer. In order to speed the analysis, a technique known as Differential Pressure Synthesis is used, which uses the principles of superposition and linearity to combine the effect of multiple small changes in head shape, without re-running the computationally intensive simulation.

Analysis
This results in a large database of peturbed head-shapes and their resultant HRTFs. The analysis of this large amount of data is the purpose of this project.