Global HRTF Interpolation via Learned Affine Transformation of Hyper-conditioned Features

TL;DR

This paper introduces a deep learning approach that uses learned affine transformations of hyper-conditioned features to interpolate and generalize HRTFs across different spatial datasets, improving efficiency and accuracy in binaural audio rendering.

Contribution

A novel deep learning architecture that enhances HRTF interpolation and generalization across datasets with varying spatial sampling distributions.

Findings

Improved cross-dataset HRTF interpolation accuracy.

Robust reconstruction of target HRTFs from sparse data.

Enhanced model generalizability across coordinate systems.

Abstract

Estimating Head-Related Transfer Functions (HRTFs) of arbitrary source points is essential in immersive binaural audio rendering. Computing each individual's HRTFs is challenging, as traditional approaches require expensive time and computational resources, while modern data-driven approaches are data-hungry. Especially for the data-driven approaches, existing HRTF datasets differ in spatial sampling distributions of source positions, posing a major problem when generalizing the method across multiple datasets. To alleviate this, we propose a deep learning method based on a novel conditioning architecture. The proposed method can predict an HRTF of any position by interpolating the HRTFs of known distributions. Experimental results show that the proposed architecture improves the model's generalizability across datasets with various coordinate systems. Additional demonstrations show that the model robustly reconstructs the target HRTFs from the spatially downsampled HRTFs in both quantitative and perceptual measures.