Providing users with a personalized HRTF set is paramount for an immersive VR experience, free from localization errors and inside-the-head sound perception.
However, direct acoustic measurement of the user's HRTF requires specialized apparatuses and is often strenuous and expensive.
We present a hybrid approach to HRTF modeling which requires only 3 anthropometric measurements and an image on the pinna contours.
A prediction algorithm based on variational autoencoders synthesizes a pinna response from its contours, which is used to filter a measured head-and-torso response.
The ITD is then manipulated to match that of a HUTUBS dataset subject minimizing the predicted localization error.
The performances are evaluated using spectral distortion and a perceptual localization model.
While the latter is inconclusive regarding the efficacy of the structural model, the former metric shows promising results with encoding HRTF datasets.