The head-related transfer function (HRTF) describes the effect the human body has on an impinging sound source. Modeling of these effects has become important due to the time-consuming measurement process and size of the HRTF data. The models reduce the HRTF into a set of parameters, making it more accessible and computationally efficient for real-time applications. The models discussed are divided into two categories: signal and structural models.

Signal models reduce irrelevant data within the HRTF to parameterize the transfer function. Minimum-phase of HRTF is used as a basis to approximate the interaural time differences. A more mathematical approach using transform decompositions, such as principal component analysis or singular value decomposition, is most advantageous for interpolation between HRTF. Each signal model strategy achieves satisfactory results but is unable to be used for individualized listening.

Structural models apply sub-models to each anthropometric object, cascading each model to construct the HRTF. The main parts of the body associated with an HRTF, mainly the torso, head, and pinnae, affect different parts of the spectrum. The superposition of the sub-models ensures the overall design is valid for the entire spectrum. Spheres and ellipsoids have become a base model of the head. Interaural time differences are calculated using head dimensions and ear locations as parameters. An extension of the sphere approximation is the snowman model. This method uses two spheres to estimate the torso and head. Reflection paths of the torso and head are added to the previous model for a more refined approximation. Pinnae models have many strategies: delay line of filters, physical models of wave phenomenon using numerical techniques. Numerical techniques have become more realizable with modern processing power and parallel computing.

hrtf model
Figure 1. Example of structural model signal flow.

More Information