Zoom out Search Issue

ManualsBrandsContents Manualsaudio & home theatreZoom in

101

102

103

104

105

106

107

108

109

110

IEEE SIGNAL PROCESSING MAGAZINE [108] MARCH 2015

parameters [30]. One of the major challenges today to numeri-

cally model the HRTF is the very high resolution of imaging

techniques required for accurate prediction of HRTFs at high

frequencies. The required resolution of the mesh imaging

depends on the shortest wavelength, which is around 17 mm at

20 kHz [13]. Moreover, obtaining these optical descriptors

demands for the use of extremely expensive laser, MRI scanners,

and also requires highly skilled, qualified users.

PERCEPTUAL FEEDBACK

Several attempts have been carried out to personalize HRTF from

a generic HRTF database using perceptual feedback. Subjects

select the HRTFs through listening tests, where they choose the

HRTFs based on the correct perception of frontal sources and

reduced front–back reversals [13]. Listeners can also adapt to the

nonindividualized HRTF by modifying the HRTFs to suit his or

her perception. Middlebrooks observed that the peaks and notches

of HRTFs are frequency shifted for different individuals and that

the extent of the shift is related to the size of pinna [31]. Listeners

often tune the spectrum until they achieve a satisfactory and natu-

ral spatialization [13]. Other techniques involve active sensory

tuning [26] and tuning the PCA weights [32] to individualize the

HRTFs. These perceptual-based methods are much simpler in

terms of the required resources and effort compared to the indi-

vidualization methods using acoustical measurements or anthro-

pometric data. However, these listening sessions can sometimes be

quite long and result in listener fatigue.

FRONTAL PROJECTION PLAYBACK

More recently, a study by Sunder et al. [33] customized the non-

individualized HRTFs using a frontal projection headphone.

Unlike side projection of sound in conventional headphones, a

frontal projection headphone projects the sound from the front to

emulate the playback from a physical set of loudspeakers. By pro-

jecting the sound from the front, the idiosyncratic frontal pinna

spectral cues of the listener are captured inherently during the

playback [33]. It is found that the idiosyncratic high-frequency

pinna cues captured in the frontal projection headphones

response match well with the frontal HRTF cues, giving it a bet-

ter frontal perception (as shown in Figure 4). The authors of

[33] reported that the front–back reversals were reduced by

almost 50% [33] using the frontal projection headphone, thus

improving the veracity of the 3-D audio. The advantage of this

technique is that it does not require any measurements, train-

ing, or the anthropometric data of the listener. However, the

frontal projection individualization technique has been limited

to only the horizontal plane and also requires a special kind of

headphone equalization (Type-2).

As discussed previously, head tracking is important in the

virtualization process. It was found that head tracking, when

used with nonindividualized HRTFs, can improve the localiza-

tion [10]. However, head tracking primarily helps in reducing

the front–back confusions and has minimal effect in reducing

the elevation localization errors, IHL [10], and coloration

caused by nonindividualized HRTFs. Since individualization of

HRTFs can alleviate some of these limitations, it is suggested

that head tracking be used with individualized rendering.

In summary, there is a noticeable trend to achieve more and

more accurate individualization with lesser data, complexity,

and effort. However, the effect of individualization of HRTFs can

be hindered by the presence of the headphones. Hence, the

headphones have to be compensated to ensure that the spec-

trum at the eardrum has only the individualized HRTF features.

Additionally, equalization of the binaural recording itself may be

necessary in certain applications (e.g., musical recordings). The

challenges and methods of equalization for both binaural and

stereo recordings are explained in the next section.

EQUALIZATION

Headphones are not acoustically transparent as they not only

color the sound that is played from the headphone but also affect

the free-air characteristics at the ear. Typically the HPTF com-

prises the headphones transducer response and the acoustic cou-

pling between the headphones and the listener’s ears. To

compensate for the headphone response, the HPTF is first mea-

sured at the same point where the recording was carried out at

the blocked ear canal or at the eardrum [35]. The binaural

recording is then deconvolved with the HPTF to eliminate the

effect of the recording microphones and the headphone. This type

of direct equalization is also known as the nondecoupled mode of

equalization (Table 4) [36]. This method is often used when the

HPTF is measured with the same measurement setup as the

recording and particularly works well when the HPTF measure-

ment and recording are carried out on the same dummy’s head.

It is observed that, in the absence of headphone equalization,

the front–back reversals are increased and the elevation localiza-

tion is distorted [1], [13], [26]. Thus, headphone equalization is

critical to create a convincing perception of virtual sound sources.

However, headphone equalization is challenging since the HPTF

depends on individual morphology (headphone–ear coupling).

Researchers have also reported that the use of nonindividualized

equalization can reduce the externalization and the effect can be

Frequency (Hz)

Magnitude (dB/20 μPa)

100

110

120

Frontal Projection Response

Frontal HRTFs

5 kHz 16 kHz

High-

Frequency

Cues

[FIG4] A comparison of the frontal projection headphone

response and the frontal directional HRTFs measured on a

dummy’s head. (Figure used courtesy of [33].)

THE WORLD’S NEWSSTAND

THE WORLD’S NEWSSTAND