Page 124 - ITU Journal, ICT Discoveries, Volume 3, No. 1, June 2020 Special issue: The future of video and immersive media

ITU Journal: ICT Discoveries, Vol. 3(1), June 2020

◦
1.1 Directional weighting estimation in ITU-R Table 2 – Binaural summation gains computed for φ = 0 and
BS.1770-4 directional weights proposed by the authors of [4] to RG-32.
Directional weights in the Recommendation were sug- Azimuth (θ) 0 ◦ ±30 ◦ ±60 ◦ ±90 ◦ ±110 ◦ ±135 ◦ 180 ◦
Computed
gested in contributions to the Rapporteur Group, and later
levels (dB) 0.00 1.36 4.47 5.22 4.46 0.84 −8.25
disclosed by Komori et al. [4]. Although the documents do Normalised
not provide further detail on how these calculations were gains (dB) 0.00 0.39 1.29 1.50 1.28 0.24 −2.37
Proposed
made, they can be traced back to the references below.
weights (dB) 0.00 0.00 1.50 1.50 1.50 0.00 −1.50
Robinson and Wittle first performed a subjective test to in- summation law in [5]. The authors observed that the effect
vestigate loudness as a function of the orientation of the of the contralateral incidence in the response variable was
sound source. Through a series of sound pressure level larger in the listening test with naive participants, although
(SPL) measurements at the ears of the listeners, the authors the difference in binaural gains in both studies might be due
stated a binaural summation law of the form: to chance, according to their statistical analysis [8].
L le ft L right

L = g×log 2 2 g +2 g , (1) Additionally, experiments in [5, 6] were conducted in ane-
choic chambers using single channel narrowband noises as
◦
where g is a 6 dB binaural gain and L is the sound pressure stimuli, while the derived weights for |φ| < 30 were tested
level equivalent to any combination of incident sound pres- in [3, 4] with broadband content rendered to 5.1, 7.1 and
sure levels, being them diotic (L left = L right ) or dichotic 22.2 loudspeaker settings, resulting in different correlations
(L left 6= L right ) [5]. between objective measurements and subjective scores ob-
served in the test sites. It is possible that these different
A different binaural summation gain for Equation (1) was results were due to elevation effects not accounted for in
derived with the method proposed by Sivonen and Eller- the weighting scheme of Table 1. Therefore, the question
meier in an experiment with narrowband, anechoic stim- on how to model directional effects on the ITU-R loudness
uli. The experimental gain g was estimated by a mini- algorithm requires further investigation.
mization of the sum-of-squares of the errors (SSE) between
the directional loudness sensitivities (DLS) of listening test The goal of the present study was to obtain subjective data
subjects and the sensitivities computed by Equation (1). on directional effects in order to estimate a new set of bin-
Squares were summed across I azimuth angles and J repe- aural summation gains. The next sections contain a de-
titions [6]. The minimum SSE is calculated as: scription of the listening test, followed by an attempt to
reproduce the estimation that led to ITU-R BS.1770-4 and
" #
I J
2 by a new approach to the problem. The modified algorithm
min ∑ ∑ DLS i,j − L comp i (g)−L ref (g) , (2)
g was then tested against a different set of subjective data on
i=1 j=1
multichannel audio.
and L ref are levels computed with Equa-
where L comp i
tion (1), corresponding to the compared incidence, and to
the frontal incidence of reference, respectively. The study 2. LISTENING TEST
, ∀i from individual Head-Related
obtained L ref and L comp i
Transfer Functions (HRTFs) of the expert subjects in their A loudness matching test was undertaken to obtain DLS re-
listening test. A value of g ≈ 3 dB was then estimated by sponses through SPL adjustments required for equal loud-
averaging Equation (2) computations per participant. ness of sounds coming from different azimuths and eleva-
tions. For this listening test, a 22-channel electroacoustic
Authors in [4] computed the channel weighting values, or system was used to reproduce broadband pink noise test
binaural loudness summation gains, in Table 2 by comput- signals in a ITU-R BS.1116 critical listening room [10] .
ing Equation (1) with g = 3 dB and L left (θ) and L right (θ)
obtained from HRTFs of each azimuth angle θ = ϑ in the 2.1 Design
table. Based on the verification that the effect of incidence
angle on loudness is attenuated for wideband and rever- Broadband pink noise stimuli, bandlimited from 200 Hz to
berant sounds [7], the authors chose to normalize results 15 kHz, were reproduced by a 22 loudspeaker setup speci-
to 1.5 dB and approximate them in 1.5 dB steps to ensure fied as layout ‘H’ in Recommendation ITU-R BS.2051 for
backward compatibility, leading to the directional weight- advanced sound systems [9]. The layout is described in
ing gains of the ITU model summarized in Table 1. Table 3 where labels indicate bottom, middle, upper and
top loudspeakers; and their correspondent azimuths. The
However, a further study by the authors in [6] computed
time-aligned and level-equalized system was mounted in an
, ∀i obtained through
ITU-R BS.1116 standard listening room with dimensions
Equation (2) with L ref and L comp i
SPL measurements taken with a Head and Torso Simula-
7.35m length, 5.7m width, and 2.5m height [11]. Mean re-
tor (HATS), and with DLS subjective scores taken from
verberation time between 500 Hz and 1 kHz octave bands
naive participants. Minimization of the objective function
is RT 60 = 0.22s.
in Equation (2) yielded g ≈ 6 dB, closer to the binaural
102 © International Telecommunication Union, 2020