Zoom out Search Issue

IEEE SIGNAL PROCESSING MAGAZINE [107] MARCH 2015
may be obtained by analytical or numerical methods, such as
the boundary element method (BEM) or the finite element
method (FEM) [13], [26]. Other methods used include multiple
linear regressions [26], multiway array analysis [28], and artifi-
cial neural networks [26]. The inputs to these methods can be a
simple geometrical primitive [29] (e.g., a sphere, cylinder, or an
ellipsoid), a 3-D mesh obtained from a magnetic resonance
imaging (MRI) machine or laser scanner or a set of two-dimen-
sional (2-D) images [13]. An important advantage of these tech-
niques is that the relative effects of a particular morphological
element (e.g., torso, head, and pinna) and their variation with
size, location, and shape can be independently investigated [13].
Another technique used a simple customization technique,
where an HRTF is selected by matching certain anthropometric
[TABLE 3] A COMPARISON OF THE VARIOUS HRTF INDIVIDUALIZATION TECHNIQUES.
HOWTOOBTAIN
INDIVIDUAL FEATURES
TECHNIQUES PROS CONS
PERFORMANCE
AND REMARKS
ACOUSTICAL
MEASUREMENTS
INDIVIDUAL MEASUREMENTS [25],
IRCAM FRANCE, CIPIC, UNIVERSITY OF MARYLAND,
TOHOKU UNIVERSITY, NAGOYA UNIVERSITY
AUSTRIAN ACADEMY OF SCIENCES [26]
IDEAL, ACCURATE REQUIRES HIGH
PRECISION; TEDIOUS;
IMPRACTICAL FOR
EVERY LISTENER
REFERENCE FOR
INDIVIDUALIZATION
TECHNIQUES
ANTHROPOMETRIC
DATA
OPTICAL DESCRIPTORS:
3-D MESH, 2-D PICTURES [13]
BASED ON ACOUSTIC
PRINCIPLES; STUDIES
THE EFFECTS OF
INDEPENDENT ELEMENTS
OF THE MORPHOLOGY
NEED A LARGE
DATABASE; TEDIOUS;
REQUIRES HIGH-
RESOLUTION IMAGING;
EXPENSIVE EQUIPMENT;
QUALIFIED USERS
USES THE
CORRELATION BETWEEN
INDIVIDUAL HRTF AND
ANTHROPOMETRIC DATA
ANALYTICAL OR NUMERICAL SOLUTIONS:
PCA + MULTIPLE LINEAR REGRESSION [26]
FEM, BEM [26], [13], MULTIWAY ARRAY ANALYSIS
[28], ARTIFICIAL NEURAL NETWORK [26]
STRUCTURAL MODEL OF HRTFs [13], HRTF
DATABASE MATCHING [30]
LISTENING/TRAINING SELECTION FROM NONINDIVIDUALIZED
HRTF [13], FREQUENCY SCALING [31]
EASY TO IMPLEMENT;
DIRECTLY RELATES TO
PERCEPTION
TAKES TIME; REQUIRES
REGULAR TRAINING;
CAUSES FATIGUE
OBTAINS THE BEST
HRTFs PERCEPTUALLY
TUNE MAGNITUDE SPECTRUM [13], ACTIVE
SENSORY TUNING [26], PCA WEIGHT TUNING [32]
SELECT CEPSTRUM PARAMETERS [34]
PLAYBACK MODE FRONTAL PROJECTION HEADPHONE [33] NO ADDITIONAL
MEASUREMENT,
LISTENING TRAINING
NEW STRUCTURE;
NOT APPLICABLE TO
NORMAL HEADPHONES;
TYPE-2 EQUALIZATION
AUTOMATIC
CUSTOMIZATION,
REDUCED FRONT–BACK
CONFUSIONS
NONINDIVIDUALIZED
HRTF
GENERALIZED HRTF [1] EASY TO IMPLEMENT NOT ACCURATE; POOR
LOCALIZATION
NOT AN
INDIVIDUALIZATION
TECHNIQUE
[FIG3] (a) Human ears act as a natural filter in physical listening. (b) The natural HRTF filter is modeled by a digital filter using various
individualization techniques. (c) Note the vast variation of the HRTF spectrum at high frequencies of the various subjects taken from
the Center for Image Processing and Integrated Computing (CIPIC) database and the Massachusetts Institute of Technology’s Knowles
Electronic Manikin for Acoustic Research (KEMAR) dummy head database [26]. This is due to the idiosyncratic nature of the pinna.
(a)
(b) (c)
Sound
Source
Sound
Source
Sound at Eardrum
with Individual
Pinna Features
Sound at Eardrum
with Individual
Pinna Features
Natural Filter
Individual
Parameters
HRTF
HRTF
Individualization
Process
Subject 131
Subject 133
Subject 156
KEMAR
10
2
10
3
10
4
Magnitude
Frequency (Hz)
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
q
q
M
M
q
q
M
M
q
M
THE WORLD’S NEWSSTAND
®
Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page
q
q
M
M
q
q
M
M
q
M
THE WORLD’S NEWSSTAND
®