CIPIC Interface Laboratory: Research

[CIPIC spatial sound research activities]

[University of California, Davis home page]

Spatial Sound Research

What are our goals?

The basic goal of our research is to develop cost-effective methods for synthesizing fully three-dimensional spatial sound. Our approach is based on measuring, understanding, and modeling the effects of the human body on incident sound waves. To that end, we have developed a unique facility for high-spatial-resolution HRTF measurement, a variety of tools for HRTF analysis and display, and a family of physically-based structural HRTF models that can be customized to individual listeners.

Measuring the HRTF of a KEMAR manikin

Support for our research comes from the National Science Foundation and from several industrial affiliates. We are currently collaborating with colleagues at the University of Maryland and Duke University on an NSF-sponsored research program whose goal is to use computer vision techniques to obtain accurate models of the body, which will in turn be used to provide the boundary conditions for computing individualized HRTFs numerically.

What is the problem?

The sizes and shapes of torsos, heads and particularly the pinnae vary substantially from person to person. Since these factors contribute significantly to the HRTF, individualized or custom HRTF are needed to obtain a faithful perception of spatial location.

Size and shape of pinnae vary from person to person

One of the problems that we are currently addressing is the development of a parameterized HRTF model that can be easily customized for individual listeners. By providing the acoustic cues with which the listener is familiar, such a model will produce significantly more realistic and convincing spatial sound.

What is our approach?

Our research is based on the belief that the HRTF can be modeled by a physically-based model employing a small number of free parameters. We anticipate that these parameters can be adapted or customized to individual listeners by correlation with a small number of properly chosen anthropometric measurements.

[Measuring the HRTF of an isolated pinna]

Measuring the response of an isolated pinna Left: the measurement system Right: closeup view of a pinna mold

Based on these premises, we are proceeding to develop and validate HRTF models using a combination of the physical and mathematical approaches. Since our models have to provide the proper sound localization cues to human listeners, we perform psychoacoustical experiments to validate their performance.

What have we accomplished?

First, we have shown that structural models can be effective in synthesizing spatial sound ( Brown and Duda 98). We have shown that a spherical model of the head provides strong range cues for close sources (Duda and Martens 98), and that the parameters for this model can be accurately estimated from anthropometry (Algazi, Avendano and Duda 01). We have demonstrated that an ellipsoidal head model can account for the variations of the interaural time difference with elevation (Duda, Avendano and Algazi 99), and that an ellipsoidal torso model can provide additional elevation cues (Avendano, Algazi and Duda 99). Furthermore, this modeling work has revealed the existence of previously unrecognized, low-frequency binaural cues for elevation (Algazi, Avendano and Duda 01). Finally, we have shown that the complex behavior of the contralateral pinna need not be reproduced in detail, but can be effectively approximated by applying head shadow and delay to the transfer function for the ipsilateral pinna (Avendano, Duda and Algazi 99). In general, our progress is documented in more than fifteen conference and journal publications.

We have also built a measurement facility that has enabled us to obtain accurate, high-resolution HRTF measurements. Small loudspeakers are attached at 5^o intervals in azimuth around a computer-controlled rotating hoop. The hoop can be rotated about the interaural axis in 5.625^o increments in elevation over a range of 270^o. The HRTF data is collected by measuring the head-related impulse responses (HRIRs), either using Golay-code based hardware (Crystal River Engineering's Snapshot^TM system) or using maximum-length sequences generated by Tucker-Davis Technology's System II.

Measuring the HRTF of a human subject

We have used this facility to measure HRTFs for more than 50 different subjects. These measurements are being organized as an HRTF database that includes anthropometric data extracted from digital photographs. This database, which will soon be made available to interested researchers, is providing us with the information needed for systematic study of individual differences in HRTFs. We believe that this will provide us with the basis for replacing the time-consuming process of measuring HRTFs acoustically with the ability to compute HRTFs from imagery.