Zoom out Search Issue

ManualsBrandsContents Manualsaudio & home theatreZoom in

111

112

113

114

115

116

117

118

119

120

IEEE SIGNAL PROCESSING MAGAZINE [111] MARCH 2015

are rendered in a manner so as to

recreate a natural sound environ-

ment. Modeling the acoustics of the

natural sound environment by add-

ing the correct amount of early

reflections and reverberation also

helps in enhancing the perception

of the sound environment as well as

veridical distance, which is critical

for natural listening. Moreover, a suitable individualization

technique has to be applied to the directional sources such that

the rendered sound scenes played over headphones are maxi-

mally tailored for the individual listener. Meanwhile, the use of a

robust equalization technique can significantly reduce the

adverse coloration of the source. Finally, the influence of the

head movements on the rendered sound can be taken into

account by incorporating head tracking in virtualization.

In general, natural sound rendering requires both the spatial

and timbral quality of the reproduced sound to be realistic. For

digital media content that contains plenty of spatial cues (e.g.,

movies, games), all five techniques reviewed are important in

creating a sense of immersiveness. For other content, where the

timbral quality is of utmost importance (e.g., music record-

ings), a subset of the techniques (e.g., individualization, equal-

ization) are sufficient.

SUBJECTIVE EXPERIMENTS

Subjective experiments were carried out to validate the

reviewed natural sound rendering system by comparing it with

the conventional stereo playback system. A total of 18 subjects

(15 males and three females), who were all between 20 and 30

years old, participated in this listening experiment. None of the

subjects reported any hearing loss. The test was conducted in a

semianechoic listening room at Nanyang Technological Univer-

sity (NTU) in Singapore. The two systems of headphone listen-

ing tested in this experiment were:

■ Conventional stereo system: The materials are directly

played back over headphones without any processing.

■ Natural sound rendering system: The signal processing

techniques introduced in the article were applied to the

audio content.

In this study, we chose PAE as the sound scene decomposition

method since our primary interest lies in movie and gaming

audio content that contains the individual sound sources and the

sound environment [21]. Individualization is carried out by fron-

tal projection headphones since it inherently embeds the personal

pinna cues during playback and does not require any individual

acoustical experiments, anthropometric data, or training [33]. To

fully exploit the frontal projection in the natural sound rendering,

we have developed a new four-emitter headphone [39] that

houses a frontal emitter and a conventional side emitter in each

ear cup of the headphone [33]. In the virtualization process, the

frontal emitters are used to render the directional sources, while

all the emitters (both frontal and side) are used to render the

sound environment. Type-2 equalization is applied to the frontal

emitters for source rendering [33],

and DF equalization is used to ren-

der environment signals over all the

emitters. Head tracking has not

been incorporated in this system.

The stimuli used in this experi-

ment were binaural (a motorcycle

in a storm and a bee at a waterfall),

movies (Brave, Prometheus), and

gaming tracks (Battlefield 3), which contain numerous spatial

cues. Each track was played back using the two headphone

playback systems tested in this article. The tracks correspond-

ing to the two systems were named “A” and “B” and played

back in a random order. The listening tests were conducted in a

double-blind manner, where both the experimenter and the

subjects were unaware of the order of the stimuli. In this

experiment, four audio quality measures were considered to

evaluate the performance of the two systems. Their descrip-

tions are:

1) Sense of direction: How clear or distinct are the per-

ceived directions of the sound objects?

2) Externalization: How clear is the stimulus perceived out-

side the head?

3) Ambience: How clear and natural is the perceived ambi-

ence of the sound environment?

4) Timbral quality: How realistic is the timbral quality of

the sound?

Subjects were asked to give scores for the four measures for

each of the two tracks “A” and “B.” The scores were based on a

0–100 scale where subjects rated 0–20 (Bad), 21–40 (Poor),

41–60 (Fair), 61–80 (Good), and 81–100 (Excellent). Finally,

the subjects were also required to indicate their overall prefer-

ence for the two tracks by selecting one of the following three

choices: “Prefer A,” “Not sure,” or “Prefer B.” To carry out this

experiment, a graphical user interface was created, which ran-

domized the order of the stimuli and automatically stored the

responses of the subjects in a file.

The responses of the subjects were analyzed for both sound

rendering systems. Figure 6 shows the overall comparison

between the two systems in terms of the mean opinion score

(MOS), scatter plot, and the overall preference of the subjects.

In (a), the MOS of the four measures for the two systems were

computed across all 18 subjects and five stimuli. While the

MOS for the conventional stereo system for all the measures

were around 60, the natural sound rendering system per-

formed much better with an MOS of over 70. An analysis of

variance (ANOVA) was conducted to generalize these results to

the whole population of listeners. The p-values were found to

be very small (<< 0.01) for all measures, indicating that the

improved performance of the natural sound rendering system

over the conventional stereo system is statistically significant.

The scatter plot in Figure 6(b) implies that most of the sub-

jects gave a higher score for the natural sound rendering sys-

tem for all the four measures. The overall preference of the

subjects across all the five tracks is shown in Figure 6(c). The

IN GENERAL, NATURAL

SOUND RENDERING REQUIRES

BOTH THE SPATIAL AND

TIMBRAL QUALITY OF

THE REPRODUCED SOUND

TO BE REALISTIC.

THE WORLD’S NEWSSTAND

THE WORLD’S NEWSSTAND