Further observations on a principal components analysis of head-related transfer functions

Abstract

Humans can externalise and localise sound-sources in three-dimensional (3D) space because approaching sound waves interact with the head and external ears, adding auditory cues by (de-)emphasising the level in different frequency bands depending on the direction of arrival. While virtual audio systems reproduce these acoustic filtering effects with signal processing, huge memory-storage capacity would be needed to cater for many listeners because the filters are as unique as the shape of each person’s head and ears. Here we use a combination of physiological imaging and acoustic simulation methods to confirm and extend previous studies that represented these filters by a linear combination of a small number of eigenmodes . Based on previous psychoacoustic results we infer that more than 10, and as many as 24, eigenmodes would be needed in a virtual audio system suitable for many listeners. Furthermore, the frequency profiles of the top five eigenmodes are robust across different populations and experimental methods, and the top three eigenmodes encode familiar 3D spatial contrasts: along the left-right, top-down, and a tilted front-back axis, respectively. These findings have implications for virtual 3D-audio systems, especially those requiring high energy-efficiency and low memory-usage such as on personal mobile devices.

Pinna-related transfer functions and lossless wave equation using finite-difference methods: Verification and asymptotic solution

“…Mokhtari et al [318] extended the study in [323] and performed another PCA investigation and a discrete cosine transform of HRTF magnitudes by employing FDTD (∆X = 2 mm) simulations coupled with a near-to-far integral approximation for 38 ears acquired from 19 adults.…”

An overview of machine learning and other data-based methods for spatial audio capture, processing, and reproduction

“…Traditionally, principal component analysis (PCA) has been the technique of choice for dimensionality reduction in HRTF datasets, leading to many interesting observations and experiments that evaluate the impact of different eigenmodes on spatial audio perception [188]. Further improvements in PCA-based HRTF modeling and customization were presented in [189,190]. HRTFs were synthesized in [191] using a sparse combination of a subject's anthropometric features.…”

Applied Methods for Sparse Sampling of Head-Related Transfer Functions

“…A recent overview of HRTF interpolation by PCA is given by Xie [4]. Various studies had found PCA to be effective in HRTF representation and estimation [13][14][15]. Existing PCA applications to HRTF interpolation involve non ear-aligned HRTFs.…”