Contribution
A 3D morphable head and a pinna model for HRTF Individualization
* Presenting author
Abstract:
Head-Related Transfer Functions are required for personalized binaural headphone reproduction and are highly individual due to anatomical differences between the subjects. Data driven individualization methods use easy and more accessible representations like ear images or video input to predict the listener’s HRTF. To learn the relationship between an image of the head and the HRTF, a large data base is required. For that reason, we first acquired a representative database of 110 Europeans by equally sampling 11 age groups, ranging from 18 to 73 years as well as males and females. We used this database to build a 3D morphable head and a pinna model. Such parametric model allows us to generate an arbitrary amount of new heads and ears within the shape space spanned by the database. For building the morphable model, we used As-Rigid-As-Possible (ARAP) technique to deform meshes during the registration, followed by a Singular Value Decomposition (SVD) to span shape space. We will present the generation and validation of the 3D morphable model along with newly generated heads that are intended to be used to generate data for learning based HRTF individualization.