Dataset Augmentation and Dimensionality Reduction of Pinna-Related Transfer Functions

TL;DR

This paper explores how dataset augmentation and principal component analysis can effectively model individual variations in head-related transfer functions for personalized binaural audio synthesis.

Contribution

It demonstrates that PCA models trained on an augmented WiDESPREaD dataset outperform those trained on original data in reducing dimensionality of PRTFs.

Findings

WiDESPREaD dataset improves PCA performance

Model trained on WiDESPREaD outperforms others

Effective dimensionality reduction of PRTFs achieved

Abstract

Efficient modeling of the inter-individual variations of head-related transfer functions (HRTFs) is a key matterto the individualization of binaural synthesis. In previous work, we augmented a dataset of 119 pairs of earshapes and pinna-related transfer functions (PRTFs), thus creating a wide dataset of 1005 ear shapes and PRTFsgenerated by random ear drawings (WiDESPREaD) and acoustical simulations. In this article, we investigate thedimensionality reduction capacity of two principal component analysis (PCA) models of magnitude PRTFs, trainedon WiDESPREaD and on the original dataset, respectively. We find that the model trained on the WiDESPREaDdataset performs best, regardless of the number of retained principal components.