Binaural reproduction and virtual auditory display  537

Here, the final result is irrelevant to the transmission response of the microphone.Therefore, the influence of M1 can be cancelled using either a microphone with an ideal transmission response or the same microphone used for binaural recording (or HRTF measurements). Then, Equations (11.7.10) and (11.7.13) become

HpTF is vital for headphone equalization. Numerous studies have measured the HpTFs (Mϕller et al., 1995b; Pralong and Carlile, 1996; Kulkarni and Colburn, 2000; Rao and Xie, 2006). Results vary across different studies. Some underlying principles can be determined by comparing headphone structures. A circumaural headphone with a larger volume of the cavity causes less compressive deformation of the pinnae; therefore, HpTFs among different headphone placements slightly differ. This type of headphone also exhibits reasonable repeatability in individualized HpTF measurements. A supra-aural headphone or a circumaural headphone with a smaller cavity volume makes the pinnae prone to compression deformation. The deformation varies with each measurement. In such a situation, the measured HpTF differs with headphone placement, especially at high Q-value spectral peaks and notches. An in-ear headphone is inserted into the ear canal to form a one-dimensional transmission system, and the influence of the compressive deformation of the pinnae is avoided. This problem should be noted in the choice of headphones.

Moreover, because of individual differences in the external ear, the HpTFs of circumaural headphones differ among individuals (Pralong and Carlile, 1996). Individualized differences in HpTFs mainly occur at frequencies above 6 kHz, similar to that of the spectral features of individualized HRTFs (Section 11.3.2). In fact, with circumaural headphones, the ear is entirely surrounded by a cushion. Therefore, the response of this type of headphone captures many of the external ear filtering effects similar to those in HRTFs is not surprising. Ideally, individualized HpTFs should be employed in headphone equalization to replicate binaural signals at the eardrum accurately. Otherwise, nonindividualized HpTF equalization likely impairs the localization cues in individualized HRTFs.

11.7.2  Some problems with binaural reproduction and VAD

Binaural reproduction can theoretically replicate the pressures at eardrums caused by a target source and thus recreate a perceived virtual source in a three-dimensional space. However, numerous experimental results indicate that subject-dependent directional errors or distortions generally exist. Examples of such errors include the following:

1.Reversal error (front-back or back-front confusion) A virtual source intended for the front hemisphere is perceived at a mirror position in the rear hemisphere or less frequently the reverse. In some instances, confusion arises regarding upand down-source positions. Such confusion is termed up-down or down-up confusion.

2.For elevation errors, the angle of a virtual source in the front median plane is typically elevated to higher positions.

3.In in-the-head localization or intracranial lateralization during headphone presentation, perceived virtual sources or auditory events are often located on the surface of the head or even inside the head rather than outside the head although the cases of inside the head localization do not occur as frequently as those in conventional stereophonic or multichannel sound reproduction over headphones. Lateralization often exists in frontal target sources and leads to an unnatural hearing experience.

538  Spatial Sound

In Section 1.6, binaural cues (ITD and ILD) cannot determine the source direction completely, and they identify the cone of confusion where the source is located. The high-fre- quency spectral cue caused by the pinna and dynamic cue caused by head turning contributes greatly to front–back and vertical localization. In static binaural reproduction, the dynamic cue is omitted. In this case, high-frequency spectral cues are important. However, high-fre- quency spectral cues vary among individuals. Recording with an unmatched artificial head or binaural synthesis with unmatched HRTFs leads to incorrect high-frequency spectral cues in reproduction (Wightman and Kistler, 1989b; Mϕller et al., 1996). For example, Wenzel et al. (1993) showed that the front-back and up-down confusion rates of headphone reproduction, where representative (nonindividualized) HRTFs are used, increase from 19% to 31% and 6% to 18% compared with the those in free-field real source localization, respectively. Moreover, some other studies have indicated that the absence of headphone equalization, headphone equalization with nonindividualized HpTFs, and some types of headphones (e.g., supra-aural headphones) may impair the high-frequency spectral cue in reproduction (Pralong and Carlile, 1996). Errors at each stage of HRTF and HpTF measurements and signal processing cause similar problems. All of the above factors are possible reasons for the directional distortion of virtual sources in static binaural reproduction.

The errors introduced in binaural signal recording/synthesis and reproduction stages, including individualized (or matched) binaural recording, HRTF-based synthesis, and HpTF equalization, should be minimized to eliminate or reduce the reversal and elevation errors of the perceived virtual sources in headphone reproduction. However, ensuring accuracy at each stage is difficult because high-frequency spectral information is sensitive to various errors. In practice, all methods can improve the performance of static binaural reproduction partly rather than completely. A final and efficient method is to use dynamic binaural synthesis and reproduction. Introducing dynamic localization cues in binaural reproduction alleviates dependence on spectral cues for front–back and vertical localization because of redundancy in localization cues.

Many researchers agreed that lateralization is caused by incorrect spatial information on both ears during sound reproduction (Plenge, 1972, 1974). Binaural pressures and spatial information errors originate from numerous sources. Similar to the case of reversal and elevation errors, errors introduced in binaural signal recording/synthesis and reproduction stages may cause lateralization in binaural reproduction (Durlach et al., 1992). Therefore, an accurate replication of pressure at the eardrum is important for externalization (Hartmann and Wittenberg, 1996). The absence of dynamic cues is also an important reason for lateralization in static binaural reproduction (Loomis et al., 1990; Durlach et al., 1992; Wenzel, 1996; Zhang and Xie, 2013).

Many studies have emphasized that environmental reflection is essential for externalization (Durlach et al., 1992). In addition to direct sound, reflection is crucial for spatial hearing. As mentioned in Sections 1.6.6, environmental reflection is the key to distance perception. Free-field HRTFs (or HRIRs) are used in the preceding discussion on binaural synthesis; therefore, the resultant binaural signals only contain direct sound without environmental reflection. HRIRs can be replaced by binaural room impulse responses (BRIRs, Section 1.8.3) in binaural synthesis to eliminate the effect of lateralization. BRIRs can be obtained not only from measurements through an artificial head or a human subject but also from binaural room acoustic modeling or artificial reverberation algorithms (Section 7.5).Binaural signals with environmental reflections can also be directly recorded using an artificial head or a human subject. The results of the psychoacoustic experiment indicate that binaural synthesis with several preceding-order early reflections is sufficient to externalize auditory events in reproduction (Begault et al., 2001).