The Use of Head-and-Torso Models for Improved Spatial Sound Synthesis

This paper concerns the use of a simple head-and-torso model to correct deficiencies in the low-frequency behavior of experimentally measured head-related transfer functions (HRTFs). This so-called "snowman" model consists of a spherical head located above a spherical torso. In addition to providing improved low-frequency response for music reproduction, the model provides the major low-frequency localization cues, including cues for low-elevation as well as high-elevation sources. The model HRTF and the measured HRTF can be easily combined by using the phase response of the model at all frequencies and by "cross-fading" between the dB magnitude responses of the model and the measurements. For efficient implementation, the exact snowman HRTF is approximated by two time delays and two first-order IIR filters. Because the poles are independent of the location of the virtual source, this supports a simple real-time implementation that allows for arbitrarily rapid head and source motion.