Elsevier

Journal of Sound and Vibration

Volume 329, Issue 19, 13 September 2010, Pages 4093-4106
Journal of Sound and Vibration

A hybrid algorithm for selecting head-related transfer function based on similarity of anthropometric structures

https://doi.org/10.1016/j.jsv.2010.03.031Get rights and content

Abstract

As the basic data for virtual auditory technology, head-related transfer function (HRTF) has many applications in the areas of room acoustic modeling, spatial hearing and multimedia. How to individualize HRTF fast and effectively has become an opening problem at present. Based on the similarity and relativity of anthropometric structures, a hybrid HRTF customization algorithm, which has combined the method of principal component analysis (PCA), multiple linear regression (MLR) and database matching (DM), has been presented in this paper. The HRTFs selected by both the best match and the worst match have been applied into obtaining binaurally auralized sounds, which are then used for subjective listening experiments and the results are compared. For the area in the horizontal plane, the localization results have shown that the selection of HRTFs can enhance the localization accuracy and can also abate the problem of front–back confusion.

Introduction

Defined as “the ratio of the Fourier transform of the signal at the listener's eardrum to that at the center of the listener's head with the listener absent”, head-related transfer function (HRTF) and its corresponding impulse response, head-related impulse response (HRIR), are essential components of many approaches to binaurally based spatial audio synthesis. They describe the changes in the sound wave as it propagates from a spatial sound source to the human eardrums [1]. The acquisition of accurate binaural HRTFs is crucial to the generation of 3D sound. Because of the individual difference of anthropometric shape and size, HRTF varies with frequencies, directions and subjects [2]. Unfortunately, it is difficult and time consuming to measure HRTF, and is not feasible to obtain the binaural HRTFs for an arbitrary listener. Therefore, how to individualize HRTF fast and effectively becomes an opening problem.

In the recent years, more and more researchers have concentrated on the individualization of HRTF. Besides the direct measuring method, calculating by boundary element method (BEM) is also a way that can obtain good accuracy [3]. However, these two types of methods are very time consuming and are difficult to implement in the study of auralization. Some researchers tried to find the relationship between the anthropometric structures of the subject and the corresponding HRTFs, from which some simpler ways could be found to predict the personalized HRTFs. Several types of prediction algorithms have been brought forward, such as database matching [4], principal component analysis [5], structure modeling [6], [7] and other statistical approaches [8], [9]. Some of these approaches are faster while others are more accurate [10]. However, in our study on real-time auralization for room acoustic modeling, the binaural information is necessary for an arbitrary listener. To save computation time we need an improved individualization approach, which is effective as well as accurate. Since the direct measurement for each listener is quite time consuming and not practical, we hope to provide well-performing HRTFs of a listener based on just a few measurements on him. The basis of our method includes two aspects: one is that there is similarity in various persons’ anthropometric structures; another is that only part of the anthropometric parameters are crucial to the spatial hearing.

For simplicity and effectiveness of database matching method, we think it can be improved as a practical approach for our applications in auralization. To select HRTFs for any listener from a measured database, we present a hybrid algorithm that combines the method of principal component analysis, multiple linear regression analysis and database matching. First, we use principal component analysis to decompose HRTFs and extract the characteristic parameters. Then we use multiple linear regression to analyze the relationship between HRTFs and the anthropometric parameters and to find the reference parameters. Last, we use database matching algorithm to find the closest HRTFs for the listener. The algorithm has been tested by subjective localization experiments using auralized sounds produced by the convolution of dry sound and the selected HRTFs. In our current research, only the horizontal plane data are used for database matching and listening tests. The detailed algorithm will be presented in the next section.

Section snippets

Methodology

Since there are a number of anthropometric parameters related to human's hearing, the core of the hybrid algorithm is to find the most crucial anthropometric parameters (as reference parameters). Once they have been found, the main job that needs to be done in applications is to measure them and use them for database matching in order to find the best matched HRTFs.

The structure of the algorithm is shown in Fig. 1. Firstly, the direction transfer function (DTF) can be calculated from the

Selection of HRTF

The CIPIC database we used contains HRIRs of 45 subjects and 43 of them have the measured anthropometric parameters, along with some other information about the subjects [17]. The anthropometric information in the database consists of 27 measurements per subject—17 for the head and the torso (x1x17) and 10 for the pinna (d1d8, θ1, θ2). Considering there may be a relation among the anthropometric structures and spatial directions, we select 35 subjects that have completed anthropometric

Conclusions

The similarity of human anthropometric structures makes it possible to individualize HRTF through anthropometric parameters. In this paper we have presented a hybrid database matching method, which combines the method of principal component analysis, correlation analysis and multiple linear regressions. By the use of CIPIC database, we select 13 parameters from the provided 27 parameters and use them as the reference parameters for database matching. The performance of the method has been

Acknowledgements

This project was supported by Natural Science Fund of Shaanxi Province of PRC and Program for New Century Excellent Talents in University. Thanks must go to those who have participated in our measurement and localization experiments. The authors also appreciate the reviewers’ constructive suggestions.

References (19)

  • J.P. Blauert

    Spatial Hearing

    (1997)
  • J.C. Middlebrooks

    Individual differences in external-ear transfer functions reduced by scaling in frequency

    Journal of the Acoustical Society of America

    (1999)
  • B.F.G. Katz

    Boundary element method calculation of individual head-related transfer function. I. Rigid model calculation

    Journal of the Acoustical Society of America

    (2001)
  • D.N. Zotkin, J. Hwang, R. Duraiswami, L.S. Davis, HRTF personalization using anthropometric measurements, Proceedings...
  • C. Jin, P. Leong, J. Leung, A. Corderoy, S. Carlile, Enabling individualized virtual auditory space using morphological...
  • V.R. Algazi, R.O. Duda, R.P. Morrison, D.M. Thompson, Structural composition and decomposition of HRTFs, Proceedings of...
  • A. Kulkarni et al.

    Infinite-impulse-response models of the head-related transfer function

    Journal of the Acoustical Society of America

    (2004)
  • B.R. Algazi et al.

    Approximating the head-related transfer function using simple geometric models of the head and torso

    Journal of the Acoustical Society of America

    (2002)
  • S. Fontana, A. Farina, Y. Grenier, A system for rapid measurement and direct customization of head related impulse...
There are more references available in the full text version of this article.

Cited by (19)

View all citing articles on Scopus
View full text