Professional Documents
Culture Documents
Telling The Difference Preference and Prediction
Telling The Difference Preference and Prediction
The recent AES Virtual Vienna convention featured some interesting papers
on perception and sound quality evaluation. In particular an emphasis could
be found on listener preferences, clarity, and naturalness concepts in live and
reproduced sound, such as with artificial reverberation and in sound mixes.
There was also some revealing work on the effects of
different loudspeaker placements.
774 J. Audio Eng. Soc., Vol. 68, No. 10, 2020 October
FEATURE ARTICLE
J. Audio Eng. Soc., Vol. 68, No. 10, 2020 October 775
FEATURE ARTICLE
776 J. Audio Eng. Soc., Vol. 68, No. 10, 2020 October
FEATURE ARTICLE
median clarity scores, whether one looks at the first or second sets
as shown in Fig. 6. of subjective data. The second set (the inde-
In a second listen- pendent absolute ratings of different genres)
ing test, 16 stimuli seemed to benefit from a much wider range
were selected from of window sizes (50–550 ms) between the
a music archive to different frequency bands, being smaller at
represent examples the LF and HF ends than in the middle. The
of a range of genres first set of subjective results seemed better
and styles. In this predicted by a more consistent window
case listeners were size of 350–450 ms across the three bands.
expected to make Overall the results with tuned measurement
an absolute rating windows performed better than the original
of the mix clarity of fixed window metrics.
each item, without
reference to others. EFFECTS OF LOUDSPEAKER
This too resulted in a PLACEMENT
Fig. 6. Median subjective clarity scores for a listening test comparing broad range of ratings Another group of papers presented at the
multiple mixes of the same multitrack content. (Figs. 6 and 7 courtesy of across the scale convention dealt with various effects of
Parker and Fenton)t
employed. loudspeaker placement on sound quality
of the various sources in a mix, and this is The IBR analysis mentioned above is and spatial impression.
not always available. The authors’ previous shown in block form in Fig. 7. Signals are Although not strictly concerned directly
work on inter-band relationship (IBR) is divided into three frequency bands and the with sound quality, it’s instructive to note
extended—work that involved measuring crest factor in each band is measured over the results of a study that was conducted by
multi-band dynamic range to predict var- a given time window, as an indicator of Craig Cieciura and colleagues from Surrey
ious aspects of music production quality. dynamic range. These can be compared and the BBC investigating factors affect-
The tuning of window size in different fre- over time to evaluate the changes that arise. ing people’s arrangement of loudspeakers
quency bands in order to optimize clarity Previously the authors had used a fixed 400 in their homes. In “Understanding Users’
prediction is explored. ms window with 75% overlap, inspired by Choices and Constraints when Positing
Ten mixes of a multitrack recording were momentary loudness measurement, but Loudspeakers in Living Rooms” (eBrief
selected from those previously submitted in this case the length of the window was 596), the authors asked a number of partic-
for a mix competition at another university, varied between 5 and 550 ms, resulting in ipants how they would arrange between
ranging in their performance in the compe- a selection of measurements at different one and eight compact wireless loudspeak-
tition, and hoped to have differing levels of lengths that could be compared for effec- ers with the goal of enhancing their exist-
mix clarity. Identical 10-second clips were tiveness. Using a grid search procedure the ing systems. This was done in the light
then normalized in terms of their loudness. optimum combination of window sizes was of evidence that ownership of surround
These were then rated by 18 listeners in sought so as to maximize the correlation of sound systems in UK homes is low (at
relative terms in a multiple comparison test, the prediction with subjective ratings. As far around 11%), lower than soundbars and
including a low anchor stimulus. Thankfully as can be ascertained, the best correlations wireless loudspeakers (both around 17%).
the results of that test displayed a range of with subjective results seemed to depend on Consequently various studies have been
Fig. 7. Multiresolution inter-band relationship (IBR) approach used for measuring mix clarity
J. Audio Eng. Soc., Vol. 68, No. 10, 2020 October 777
FEATURE ARTICLE
interested in what has been termed “device layout geometry for ambisonic render-
orchestration”—essentially a means by ing by Lukas Gölles and his colleagues
which ad hoc wireless and smart speak- (“Influence of Horizontal Loudspeaker JAES
ers can be utilized to enhance audio scene Layout Geometry on Sweet Area Shape
reproduction. The question therefore was for Wwidened/Diffuse Frontal Sound,” offers
Open
where might people be willing to put these Paper 10369.) They point out that the
devices in their rooms? geometry of a loudspeaker array for ambi-
Three themes emerged from the study—
spatial balance and distribution, room
sonic reproduction is often adapted to the
rectangular shape of a space, even though
Access
aesthetics, and room functionality. Sadly, a perfect cube or sphere might theoreti-
The AES Journal is
but perhaps not surprisingly for the audio cally be the best layout. It may not be clear pleased to offer an Open
community, aesthetics and functionality were whether the listener is facing the long Access (OA) publishing
prized above balance and distribution, with or short side of the rectangle, or which option to its authors.
convenient surfaces such as tables and book- arrangement offers the biggest “sweet A g rowing number of
shelves inevitably being used for positioning. area.” After conducting listening experi- countries and research funding
When asked to position eight such devices ments with widened and diffused sources, bodies require Open Access to pub-
in their rooms most participants expressed looking at the listening area beyond which licly-funded research. If an AES
“trepidation” or “doubt,” with one exclaiming images collapsed into the nearest loud- paper has been made Open Access it
“I wouldn’t have that many loudspeakers in speaker, they concluded that in fact a will have the OA log o (above) next to
it and will be freely downloadable
my house!” This was even a factor for some wide rectangle layout yielded significantly
from the AES E-Library by anyone,
people with four loudspeakers. This does not better results than a long rectangle or a
even if they’re not an AES member
provide much encouragement for those that circular layout in most cases. or E-Lib rary sub scrib er. An OA
hope immersive audio systems will become In an eBrief looking at “The Influence paper can also be distributed by the
adopted widely in the consumer arena. of Loudspeaker-Listener Distance on the author and by third parties.
Continuing the theme of how many loud- Detection of Low-Bitrate Audio Coding To find further information refer
speakers are needed and where they should Artifacts” (eBrief 576), Alan Pawlak and to: http://www.aes.org /openaccess/.
be put, Kamekawa and Marui look at whether Hyunkook Lee wanted to know whether
full-range loudspeakers are necessary in all the recommended minimum distance
locations for 3D audio. (Paper 10362, “Are of two metres was actually important. A
Full-Range Loudspeakers Necessary for the distance of at least two metres from the noise in the results. Some of the papers dis-
Top Layer of Three-Dimensional Audio.”) loudspeakers seems to be specified in a cussed in this article bear out the idea that
Here they were concerned with 22.2 multi- number of standards, but the authors had if you gather a group of relatively untutored
channel sound and whether filtering the not found a clear rationale for this, or much listeners together and ask them what they
top layer speakers makes any difference. As experimental evidence. Consequently they like, or whether they can hear the differ-
they say, even though the various 3D audio compared the results of a sound quality ences between things, the results will be
system proponents hope for speakers with test on a low bitrate audio codec between inconclusive, or you will conclude that it
similar characteristics all round, “it is not headphone listening and listening at vari- doesn’t matter. The more discerning and
easy to prepare an ideal reproduction envi- ous loudspeaker distances, finding little trained the listener is, the more likely it
ronment in a consumer’s home audio…” If influence of the listening distance on the is that small differences will be noticed,
the overhead speakers could be smaller it ability to detect artifacts. Listeners were and that these differences will matter for
could help in practical terms. more affected by the reproduction system their preference. Whether the differences
Using a variety of different types of than they were by the listening distance, that highly trained listeners obsess about
program material, including classical and although the authors did note that these actually matter when it comes to imple-
traditional Japanese music, as well as voice issues are potentially listener dependent. menting consumer systems may be a ques-
and noise, the authors attempted to discover tion overridden in some cases by practical
what high-pass or low-pass filtering could CONCLUSION constraints in the home, car or life situa-
be tolerated without the listeners noticing Listener preference for one thing or tion. This is perhaps best exemplified by the
a difference from full-range content in the another, be it reverberation, mix balance, reaction of one participant in the study on
upper loudspeakers. Interestingly 75% of loudspeaker arrangements or characteris- multiple loudspeakers, mentioned above—
the listeners couldn’t distinguish any differ- tics, has always been a thorny issue. It tends “I wouldn’t have that many loudspeakers in
ence when the upper channels emitted noth- to depend who you ask, although common my house!”
ing below 400 Hz. On the other hand, high trends have been observed in a number
frequency roll-offs had more variable effects. of studies. There is some evidence that Editor’s note: the papers discussed in this
Listeners with greater 3D production experi- consumer preferences for basic aspects of article and others from the same conference
ence were more sensitive to changes in high sound quality (such as frequency response can be downloaded from the AES E-Library
at http://www.aes.org/e-lib/. AES
and low frequency response. of loudspeakers or headphones) tend to fol- members get free access to the E-Library.
There’s also a paper on loudspeaker low professional ones, just with much more
778 J. Audio Eng. Soc., Vol. 68, No. 10, 2020 October