Headphone design has traditionally focused on creating a frequency response to make commercial stereo audio sound more natural. However, because of the sensitivity of spatial hearing to frequency-dependent cues, binaural reproduction requires headphones' target spectrum to be as flat as possible. Initial attempts to equalize headphones used a naive inversion of the headphone spectrum, which degraded binaural content because the headphone transfer function (HpTF) changes each time headphones are re-seated. Many different algorithms have been proposed to improve binaural equalization, each of which has been tested over a limited sample of HpTFs. The present study gathered 1550 HpTFs from different institutions into a single dataset for large-scale comparisons of equalization algorithms. A numerical metric was designed to quantify auditory perception of spectral coloration from 'ringing' peaks in the post-equalization HpTF. Using this metric, eight of the most prominent equalization methods have been compared over the aggregate HpTF dataset. High-shelf regularization is shown to outperform all other equalization techniques using either individualized or averaged input spectra. In addition, high-shelf regularization without individual measurements gives less average coloration than direct inversion using individualized equalization.

Comparison of distortion products in headphone equalization algorithms for binaural synthesis

Geronazzo M.
2021

Abstract

Headphone design has traditionally focused on creating a frequency response to make commercial stereo audio sound more natural. However, because of the sensitivity of spatial hearing to frequency-dependent cues, binaural reproduction requires headphones' target spectrum to be as flat as possible. Initial attempts to equalize headphones used a naive inversion of the headphone spectrum, which degraded binaural content because the headphone transfer function (HpTF) changes each time headphones are re-seated. Many different algorithms have been proposed to improve binaural equalization, each of which has been tested over a limited sample of HpTFs. The present study gathered 1550 HpTFs from different institutions into a single dataset for large-scale comparisons of equalization algorithms. A numerical metric was designed to quantify auditory perception of spectral coloration from 'ringing' peaks in the post-equalization HpTF. Using this metric, eight of the most prominent equalization methods have been compared over the aggregate HpTF dataset. High-shelf regularization is shown to outperform all other equalization techniques using either individualized or averaged input spectra. In addition, high-shelf regularization without individual measurements gives less average coloration than direct inversion using individualized equalization.
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/11390/1210428
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 0
  • ???jsp.display-item.citation.isi??? ND
social impact