Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

ARTICLE

...................................

Discomfort estimation for aircraft cabin noise using linear


regression and modified psychoacoustic annoyance approaches
Yu Huang,a),b) Bingcong Lv, Ke Ni, and Weikang Jianga)
Institute of Vibration, Shock and Noise, School of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai, 200240, China

ABSTRACT:
Appropriate sound quality models for noise-induced discomfort are necessary for a better acoustic comfort design in
the aircraft cabin. This study investigates the acoustic discomfort in two large passenger aeroplanes (i.e., planes A
and B). We recorded the noise at 21 positions in each aircraft cabin and selected 42 stimuli ranging from 72 to
81 dB(A) during the cruising flights. Twenty-four participants rated the noise discomfort by the absolute magnitude
estimation method. The discomfort values in the middle section of the aircraft cabin are 10% points higher than in
the front or rear section. The discomfort magnitude was dominated by loudness and influenced by roughness and
sharpness. A multiple linear (MA) discomfort model was established, accounting for the relationship between the
discomfort and sound quality metrics (i.e., loudness, sharpness, and roughness). The MA model estimated noise dis-
comfort better than the Zwicker and other (i.e., More and Di) psychoacoustic annoyance (PA) models. We modified
the coefficients of independent variables in the formulations of Zwicker, Di, and More PA models, respectively,
according to the present experimental results. The correlation coefficients between the estimated and measured val-
ues of the modified models were at least 20% points higher than the original ones.
C 2023 Acoustical Society of America. https://doi.org/10.1121/10.0020838
V
(Received 29 November 2022; revised 15 August 2023; accepted 15 August 2023; published online 2 October 2023)
[Editor: Francesco Aletta] Pages: 1963–1976

I. INTRODUCTION Multiple linear regression (MLR) is widely used for


sound quality modelling of discomfort with one or several
Acoustic comfort, as the perceived state of well-being
psychoacoustic metrics. Zhang et al. established an acoustic
and satisfaction with the acoustic conditions in an environ-
comfort model for a forklift by MLR between the dependent
ment, has attracted more and more attention in the aerospace
variable, rank scores, and independent variables loudness
industry.1–3 Good acoustic comfort in large passenger aircraft
and articulation index.20 Li and Huang developed MLR
cabins diminishes fatigue during the journey and improves
models to predict noise discomfort in micro commercial
the safety and performance of all occupants.4 Acoustic dis-
vehicles with different driving and road conditions using
comfort and annoyance have been recognised as the adverse
various SQMs such as loudness, sharpness, and roughness.19
effect of noise on comfort in previous studies, e.g., comfort
Yoon et al.21 have built sound quality models, including
studies on aircraft interior sound and vibration.5
loudness, sharpness, and roughness metrics for the noise
Empirical studies have found that acoustic discomfort
pleasant of a vehicle’s heating, ventilation, and air condi-
(or annoyance) is dominated by sound pressure level (SPL)
tioning (HAVC) system via regression.
and moderated by the sound sources depending on their dif-
Zwicker and Fastl12 proposed a psychoacoustic annoy-
ferent characteristics.6–11 Sound quality metrics (SQMs),
ance (PA) model to estimate acoustic discomfort and veri-
e.g., loudness, sharpness, fluctuation strength, roughness,
fied that it achieved excellent performance for synthetic
and tonality, have been proposed to represent subjective
noise and mechanical noise radiated from cars, air condi-
responses to various sound features and work as psycho-
tioners, and tools.10 Some studies have modified the
acoustic indices for estimating and predicting noise-induced
Zwicker PA model to improve the estimation accuracy of
discomfort.12 Studies have been spread across areas of traf-
discomfort caused by different types of noise. More investi-
fic noise,10 construction noise,13 domestic appliance
gated the influence of signal characteristics on aircraft noise
noise,14–17 vehicle noise,18,19 etc. These studies have found
discomfort (via annoyance judgements) and modified the
that noise-induced discomfort is highly dependent on acous-
Zwicker PA model by adding tonality and a series of c coef-
tic characteristics. For example, loudness and roughness
ficients for each independent psychoacoustic variable.22 Di
impacted the acoustic discomfort significantly for construc-
et al.23 studied acoustic discomfort induced by a 1000 kV
tion noise,13 whereas air conditioner and refrigerator noise
transformer low-frequency tonal noise and developed the
discomfort were strongly correlated with pitch.14
tonality factor xT in the modified PA model. A recent study
on modelling acoustic discomfort for the rotor system noise
a) of unmanned aerial vehicles (UAV) added a parameter of
Also at: State Key Laboratory of Mechanical System and Vibration,
Shanghai Jiao Tong University, Shanghai, 200240, China. impulsiveness based on the More PA model.24 From the pre-
b)
Email: yu_huang@sjtu.edu.cn viously mentioned studies, it seems that various sound

J. Acoust. Soc. Am. 154 (4), October 2023 C 2023 Acoustical Society of America
V 1963
https://doi.org/10.1121/10.0020838

quality models with different psychophysical metrics are values of noise discomfort at various recording points were
necessary for appropriately estimating different types of rated in the laboratory experiment and compared, hypothe-
noise based on subjective experimental data, either using the sising that: (1) discomfort would differ significantly among
MLR or the Zwicker and other modified PA approaches. the front, middle and rear cabin of aircraft, and (2) among
Several studies investigated the SQMs influencing different locations at the aisle, window and air conditioning
noise-induced discomfort inside aircraft cabins. Quehl found outlet. We developed and compared the sound quality mod-
that increasing loudness, roughness, and fluctuation strength els using MLR and PA models in estimating the response
would reduce comfort in the propeller and jet aircraft cabin.4 values of discomfort and propose a modified approach to
Lopes found that sharpness, loudness, and tonality illus- improve the accuracy of PA models for the aircraft cabin
trated higher correlations with evaluated annoyance than noise.
other SQMs in the propeller aircraft cabin.25 Pennig et al.26
indicated that the comfort sensation in aircraft was domi- II. METHODS
nated by SPL and significantly affected by the frequency
A. Apparatus
characteristics changing with the seat positions, i.e., front,
middle, and rear in the aircraft cabin, through variation The experiment was conducted in a semi-anechoic
dimension regression analysis. However, there has not been chamber. The noise stimuli were played via a pair of head-
a comprehensive study on acoustic discomfort for large pas- phones (Sennheiser HD600, Wedemark, Germany) driven
senger aeroplanes. A complete understanding of how differ- by a digital-to-analogue converter (RME ADI-2DAC,
ent SQMs influence discomfort inside the aircraft cabin Haimhausen, Germany) connected to a laptop (Xiaomi
during a flight allows acoustic discomfort models based on Notebook Pro 15, Shanghai, China). MATLAB (version
sound quality modelling, with the potential benefits of more R2020a, Mathworks, Natick, MA) generated and controlled
efficient comfort designs for large passenger aircraft cabins. the stimuli. An artificial head (HMS IV, Head Acoustics,
This study intends to investigate the discomfort caused Herzogenrath, Germany) was used to calibrate the stimuli.
by noise in two large passenger aeroplanes during their The connection diagram of the experiment set-up and the
cruise at different positions of the cabins. The response scene are shown in Fig. 1.

FIG. 1. (Color online) The experimental setup. (a) The calibration system connection diagram and the experimental device connection diagram. (b) The
scene of calibrating the noise samples. (c) The scene of subjective testing.

1964 J. Acoust. Soc. Am. 154 (4), October 2023 Huang et al.
https://doi.org/10.1121/10.0020838

B. Stimuli 125 and 8000 Hz at 20 dB SPL in a hearing test before the


formal experiment. None reported a history of otological or
The noise was recorded during the flights of two
neurological disorders. All study participants signed
narrow-body large passenger aeroplanes with two jet
informed consent before the experiment.
engines, i.e., Boeing 737–800 (plane A) and Airbus A320
(plane B). Plane A is 39.47 m long, 12.55 m high, 34.32 m
D. Procedure
wingspan, 3.53 m cabin width, and has a passenger capacity
from 160 to 189. Plane B is 37.6 m long, 11.76 m high, The subjective experiment was in a semi-anechoic cham-
35.8 m wingspan, 3.70 m cabin width, and has a passenger ber. Each participant heard stimuli as diotic signals and eval-
capacity from 150 to 186. The flight was explicitly organ- uated the acoustic discomfort using the absolute magnitude
ised for the acoustic measurement, with only the flight crew estimation (AME) method, i.e., using any positive numerical
and several technical testing personnel (less than seven) on value without an upper limit to rate the discomfort magni-
board. The cabins of the two aeroplanes were divided into tude. Stevens developed the magnitude estimation method to
three parts (i.e., front, middle, and rear sections). The posi- scale the subjective intensity of stimulus.27 We adopted the
tions of the measuring points are shown in Fig. 2. We set up AME method, which has been employed to obtain the dis-
21 measuring points in plane A: No.1–No.14 at the seat comfort magnitudes of noise and vibration before, to get con-
headrest with a height of 1.2 m from the floor, and tinuous “response magnitude” as a “quantitative continuum”
No.15–No.21 at the aisle with a height of 1.7 m; 21 measur- other than “response extent” as a “qualitative continuum”
ing points in plane B: No.1–No.11 at the seat headrest, that might be more suitable with the CR method.28,29 The
No.12–No.17 at the aisle, and No.18–No.21 at 10 cm far AME method would also benefit from its unconstrained use
from the air conditioning outlet. The noise at all measuring of numbers and is relatively free of biases without any limita-
points was simultaneously measured and recorded using tions or cognitive constraints.27–29
microphones (type 4189, Br€uel & Kjaer, Naerum, Before starting the experiment, study participants were
Denmark). During the test, the aeroplanes were cruising sta- provided with written instructions and practised with the
bly, the seat back was upright, and the pressure and air con- AME method until they felt confident about the formal
ditioning systems were in regular operation. experiment. The task was presented in the instructions as
Forty-two noise samples were recorded, with 21 pieces “Please give the numerical values of discomfort (i.e., your
for each aircraft cabin. We selected a 5-s stimulus from uncomfortable sense) caused by the noise stimulus” (both in
each noise sample with no voice (e.g., speaking, broadcast- English and Chinese). Each stimulus lasted 5-s, and the
ing) and no sudden abnormal sound (e.g., door opening whole test duration for each participant was around 15 min.
sound) by time and frequency spectrum analysis and exper- For each participant, the stimuli were played in independent
imenter audition. To ensure all stimuli had the same A- random orders. The stimuli could be replayed if required by
weighted SPLs as the recording samples, we calibrated all the participants. The participants had nothing to do but to
stimuli via headphones (HD 600) on the artificial head listen and report, and there was no other disturbance during
(HMS IV). The time domain and frequency spectra of six the experiment. The protocol was approved by the Human
typical stimuli, i.e., one for each representative position, Experimentation Safety and Ethics Committee of the Bio-X
are shown in Fig. 3. Research Institute at Shanghai Jiao Tong University.

C. Study participants E. Data processing


We recruited twenty-four healthy participants, 12 males The SQMs, i.e., Zwicker loudness, sharpness, rough-
and 12 females, aged from 20 to 30 (median age: 23), at ness, fluctuation strength, and tonality, of all noise stimuli,
Shanghai Jiao Tong University. They self-reported normal were calculated by the software Artemis (version 12, Head
hearing and could perceive audiometric frequencies between Acoustics, Germany). The loudness (in sone) and the 5th

FIG. 2. (Color online) Locations of the measuring points in the front, middle, and rear cabin of two aeroplanes. The index in green for sitting position near
the window, in purple for standing position in the aisle, and in blue for air conditioning outlet. The frame line bronze solid line front section, olive solid line
middle section, maroon solid line rear section.

J. Acoust. Soc. Am. 154 (4), October 2023 Huang et al. 1965
https://doi.org/10.1121/10.0020838

FIG. 3. (Color online) The time domain and frequency spectra of typical noise stimuli at front, middle, and rear sections in two planes.

percentile of the loudness (N5 , in sone) calculation were The numerical response values of each participant
both based on the ISO532 method.30 The sharpness (in should be “standardized” or “equalized” because all the sub-
acum) was calculated according to DIN 45692.31 The tonal- jective magnitudes had to be compared in the same numeri-
ity (in tuHMS) was calculated according to ECMA-418–2.32 cal range. The discomfort magnitudes were divided by the
The roughness (in asper) and fluctuation strength (in vacil) median values of individual data over all stimuli and multi-
were calculated according to the hearing model of Sottek,33 plied by 100.34 We show the “standardized” subjective
which was also adopted in recent sound quality modelling response values of discomfort caused by each of all 42 noise
studies.22,24,45 stimuli by a boxplot in Fig. 4. Discomfort values grouped by
1966 J. Acoust. Soc. Am. 154 (4), October 2023 Huang et al.
https://doi.org/10.1121/10.0020838


FIG. 4. (Color online) Subjective discomfort response values of each stimulus. , sitting position near the window, 䉱, standing position in the aisle, , air
conditioning outlet position.

various locations (i.e., the sitting position, standing position where N5 is the 5th percentile of the loudness (in a unit of
and air conditioning outlet, refer to different index colours sone), S is sharpness (in a unit of acum), R is roughness (in a
in Fig. 2) are shown to compare among different cabin sec- unit of asper), and F is fluctuation strength (in a unit of vacil),
tions (i.e., front, middle and rear, refer to different frame xS 2 and xFR 2 are two coefficients defined in Eqs. (2) and (3).12
line colours in Fig. 2) and grouped in various sections to The More PA model,
compare among different locations.  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
All statistical and regression analyses were completed PAM ¼ N5 1 þ c0 þ c1 xS 2 þ c2 xFR 2 þ c3 xTM 2 ; (4)
in the IBM SPSS Statistics (version 23, IBM, Armonk, NY).
h i
After the Kolmogorov-Smirnov test35 on the data set (i.e., 2 2
xTM 2 ¼ ð1  ec4 N5 Þ ð1  ec5 K5 Þ ; (5)
all individual response values) for each stimulus, most data
sets were normally distributed, and the rest approximated
the normal distribution. Therefore, nonparametric statistical where xTM 2 describes the effect of tonality and is defined in
tests were employed. A Friedman test was used to compare Eq. (5), K5 is the tonality exceeded 5% of the time (in a unit
the differences in response values among multiple matched of tu), and the coefficients, c0 ¼ 0.16, c1 ¼ 11.48,
groups of stimuli.36 The significance level (i.e., the p-value) c2 ¼ 0.84, c3 ¼ 1.25, c4 ¼ 0.29, c5 ¼ 5.49, all iterated by non-
was adjusted for multiple comparisons by Bonferroni linear least square method.22
correction. The Di PA model,
All subjects’ median discomfort response values for  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
each stimulus were calculated, and all 42 median discomfort PAD ¼ N5 1 þ xS 2 þ xFR 2 þ xTD 2 ; (6)
response values of 42 stimuli followed the normal distribu-
6:41
tion (p > 0.05, Kolmogorov-Smirnov test). We investigated xTD ¼ T; (7)
the relationship between discomfort response values and N5 0:52
multiple SQMs by MLR. The zero-order Pearson correlation
(i.e., the correlation between response values and all SQMs where T is the tonality (in a unit of tu), xS 2 , xFR 2 , and xTD 2
without controlling for any variable) and partial correlation are coefficients defined in Eqs. (2), (3), and (7).23
(when controlling for loudness) were performed to investi- The reported discomfort values (inter-individual median
gate the contribution of each psychoacoustic metric to value per stimulus) and estimated discomfort values by dif-
response values. We determined the proper regression equa- ferent models were normalized to 0–1 intervals to ensure the
tions considering the partial correlation coefficient and vari- results are compared in the same numerical scale,24 by the
ance inflation factor (VIF) for the multi-collinearity following equation:
problem. xi  xmin
yi ¼ ; (8)
We also adopted three PA models for noise-induced xmax  xmin
discomfort response.
Zwicker PA model, where yi is the normalized response value, xi is the response
 value of the ith noise stimulus, xmin is the minimum, and
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi xmax is the maximum value of the data set, fxi g.
PA ¼ N5 1 þ xS 2 þ xFR 2 ; (1)
( III. RESULTS
ðS  1:75Þ  0:25lgðN5 þ 10Þ; S > 1:75;
xS ¼ (2) A. Response values of discomfort caused by noise
0; S  1:75;
in two aircraft cabins
2:18 Figure 4 shows the boxplot [including “minimum,” first
xFR ¼ ð0:4F þ 0:6RÞ; (3)
N5 0:4 quartile (Q1), median, third quartile (Q3), “maximum,” and
J. Acoust. Soc. Am. 154 (4), October 2023 Huang et al. 1967
https://doi.org/10.1121/10.0020838

FIG. 5. (Color online) The acoustic dis-


comfort of stimuli in five position groups
compared among different sections of
two aeroplanes. green solid line, Groups
AI and BI, sitting position near the win-
dow of planes A and B; violet solid line
Groups AII and BII, standing position in
the aisle of of planes A and B; blue solid
line Group BIII, air conditioning outlet
of plane B. **p < 0.01, ***p < 0.001,
****p < 0.0001, Friedman test, signifi-
cance after Bonferroni correction.

outlier] of individual subjective response values of discom- plane A or B, the response values of discomfort in the mid-
fort caused by all noise stimuli in the two aeroplanes. dle section were the greatest for the usual activity positions
Twenty-eight data (i.e., the solid circle, triangle and cross- of passengers (i.e., Groups AI, AII, BI, and BII), with median
symbols in Fig. 4) with values out of the range of the first values at least 10% points higher than the front or rear sec-
quartile (Q1) minus 1.5 times the interquartile ranges (IQR) tion. The noise at the air conditioning outlet positions (i.e.,
and the third quartile (Q3) plus 1.5 times the IQR (i.e., Group BIII) in the front section of plane B caused signifi-
[Q1–1.5IQR, Q3 þ 1.5QR]) were eliminated in the further cantly greater discomfort responses than the middle (median
analysis according to commonly used outlier removal values, 15.2% points higher) and rear (median values,
criteria.37 78.4% points higher) sections.
We divided the data into five groups: Group AI, the sit- Figure 6 shows the boxplot (including “minimum,” Q1,
ting position near the window of plane A (i.e., the green median, Q3, and “maximum”) of individual discomfort
index group in Fig. 2); AII, the standing position in the aisle response values of stimuli compared among different loca-
(i.e., the purple index group in Fig. 2); BI, the sitting posi- tions at the aisle, window, and air conditioning outlet in
tion near the window of plane B (i.e., the green index group each section (i.e., the front, middle, and rear) of two aero-
in Fig. 2); BII, the standing position in the aisle (i.e., the pur- planes. From Fig. 6, the discomfort values at the window
ple index group in Fig. 2); BIII, the air conditioning outlet position (Groups AI and BI) were significantly greater than
position (i.e., the blue index group in Fig. 2). Figure 5 shows those at the aisle position (Groups AII and BII), with median
the boxplot (including “minimum,” Q1, median, Q3, and values 28.2% points higher for plane A and 11.1% points
“maximum”) of individual discomfort response values of higher for plane B. The noise at the air conditioning outlets
the five groups stimuli (i.e., Groups AI, AII, BI, BII, and BIII) (Group BIII) caused the greatest discomfort in each section
compared among different sections (i.e., front, middle, and of plane B, with median values 44.4% points higher at the
rear) of two aeroplanes. There were significant differences window position (i.e., Group BI) for the front section, 4.9%
in discomfort values among the front, middle, and rear sec- points higher for the middle, and 5.5% points higher for the
tions for each group (p < 0.01, Friedman test). In either rear. Note that the difference in the discomfort among the

FIG. 6. (Color online) The acoustic dis-


comfort of stimuli compared among dif-
ferent locations at the aisle, window and
air conditioning outlet. Bronze solid line
front section, olive solid line middle sec-
tion, maroon solid line rear section.
*p < 0.05, **p < 0.01, ***p < 0.001,
****p < 0.0001, Friedman test and
Wilcoxon test, significance after
Bonferroni correction.

1968 J. Acoust. Soc. Am. 154 (4), October 2023 Huang et al.
https://doi.org/10.1121/10.0020838

different locations in the middle section of plane B was B01-17) of the aircraft cabin noise for establishing the
insignificant (p ¼ 0.08, Friedman test, significance after model. Table III shows the correlation matrix between every
Bonferroni correction), possibly attributed to the relatively two variables (i.e., discomfort response value, A-weighted
poor acoustic comfort situation at all locations in the middle SPL, loudness, sharpness, roughness, fluctuation strength
section. and tonality) for planes A and B. The discomfort response
values and the loudness showed the dominant effect on dis-
B. The SQMs of aircraft cabin noise comfort (i.e., the highest significant positive correlation
coefficient) for each plane. Table IV lists the zero-order
The A-weighted SPLs, sound quality metrics (i.e., loud-
(i.e., the correlation between variables without controlling
ness, sharpness, roughness, fluctuation strength, and tonality),
for any variable) and partial correlation coefficients with
and median response values of discomfort caused by noise in
controlling for the loudness between the psychoacoustic
planes A and B are listed in Tables I and II, respectively. The
metrics and the response values for planes A, B, and both
median and IQR values of the previously noted parameters of
planes. When not controlling loudness: (1) the fluctuation
21 stimuli for each plane are also marked. The noise stimuli
strength values had a substantial positive correlation with
of plane B have lower average and smaller IQR values of
discomfort response values for planes A, B, and total data;
SPL, loudness, and discomfort response than Plane A. Figure
(2) the tonality values had a positive correlation with dis-
7 shows the heatmap of the median discomfort response val-
comfort response values for plane A and total data; (3) the
ues and A-weighted SPLs at each measuring point of both
sharpness and roughness values have a positive correlation
aeroplanes. From Fig. 7, the air conditioning outlet points
with discomfort response values for plane B and a negative
(marked with the symbol ) appear to have significantly
correlation with response values for plane A and total data,
higher SPL and discomfort than other positions in plane B.
respectively. However, when controlling loudness: (1) the
As the air conditioning outlet points are too close to the noise
correlation coefficients for sharpness, fluctuation strength,
source and not the usual hearing positions of routinely active
and tonality decrease as expected; (2) whereas the signifi-
people, the data of these positions (i.e., indices B18–B21 in
cant negative effect of roughness on discomfort response
Table II) would not be employed for the following sound
values emerged with increased absolute values of the nega-
quality modelling process.
tive correlation coefficients between roughness and discom-
fort for each and both planes.
C. Sound quality models of aircraft cabin noise
The sound quality models of noise discomfort response
As indicated previously, we used the data at usual were established using the stepwise multiple linear regres-
human sitting and standing positions (i.e., A01–21 and sion method between the dependent variable, discomfort

TABLE I. The A-weighted SPLs, SQMs, and median response values of stimuli for plane A.

Median response A-weighted Loudness Sharpness Roughness Fluctuation Tonality


Index values SPL (dBA) (sone) (acum) (asper) strength (vcail) (tu)

A01 105.00 74.2 36.97 1.93 2.95 0.0629 2.0787


A02 137.86 77.2 44.06 1.48 2.96 0.0746 3.3864
A03 98.46 76.6 38.63 1.58 3.18 0.0658 1.2763
A04 100.00 77.9 41.80 1.89 3.34 0.07 0.1049
A05 124.05 78.5 45.07 1.74 3.21 0.0714 1.5601
A06 123.33 80.4 47.73 1.42 3.03 0.0778 2.5537
A07 113.33 78.9 43.69 1.33 2.86 0.0805 2.791
A08 108.75 72.5 34.28 1.27 2.52 0.0626 3.2842
A09 96.32 77.0 39.83 1.62 3.18 0.0684 0.7831
A10 100.00 78.3 41.86 1.86 3.23 0.0735 0.2488
A11 125.00 80.7 50.82 1.55 3.17 0.0781 3.1543
A12 128.57 80.2 50.97 1.54 2.85 0.0857 4.3302
A13 124.05 78.7 44.60 1.35 2.81 0.0794 3.2924
A14 100.00 78.1 41.74 1.54 3.12 0.068 0.5359
A15 57.14 73.2 29.89 1.79 2.95 0.0574 0.3258
A16 65.48 74.5 32.40 1.67 3.14 0.0552 0.1669
A17 88.33 75.7 38.17 1.79 3.16 0.0577 1.0076
A18 118.25 77.2 41.72 2.04 3.30 0.0641 0.7836
A19 99.29 78.0 39.81 1.56 3.12 0.0699 0.1056
A20 100.00 76.8 35.35 1.45 2.91 0.07 0.079
A21 98.58 77.1 37.38 1.66 3.04 0.0712 0.0723
Median 100.00 77.2 41.72 1.58 3.12 0.0700 1.0076
IQR 24.75 1.96 6.68 0.31 0.23 0.0105 2.5422

J. Acoust. Soc. Am. 154 (4), October 2023 Huang et al. 1969
https://doi.org/10.1121/10.0020838

TABLE II. The A-weighted SPLs, SQMs and median response values of stimuli for plane B.

Median response A-weighted Loudness Sharpness Roughness Fluctuation Tonality


Index values SPL (dBA) (sone) (acum) (asper) strength (vcail) (tu)

B01 60.00 74.5 33.13 1.88 3.10 0.0529 0.2768


B02 90.00 75.3 35.69 2.27 3.21 0.0655 0.1046
B03 114.28 77.3 39.15 2.04 3.11 0.0736 0.2278
B04 100.00 75.5 35.53 1.98 3.06 0.0628 0.0694
B05 100.00 77.3 39.86 2.24 3.43 0.067 0.0511
B06 88.08 76.4 37.61 2.25 3.26 0.0544 0.3206
B07 100.00 74.2 34.19 2.27 3.15 0.055 0.312
B08 100.99 77.3 38.43 2.08 3.24 0.0755 0.072
B09 100.32 77.0 38.08 2.01 3.14 0.0735 0.0655
B10 100.32 76.3 39.50 2.24 3.32 0.0577 0.2099
B11 90.87 77.7 38.48 2.41 3.52 0.0575 0.0614
B12 60.77 75.9 34.51 1.92 3.19 0.0616 0.2736
B13 86.19 75.5 35.87 2.19 3.21 0.0677 0.1841
B14 100.00 77.2 39.06 2.11 3.32 0.0683 0.0793
B15 100.00 76.8 37.65 2.05 3.21 0.0743 0.078
B16 88.33 76.8 38.53 1.95 3.32 0.069 0.0588
B17 82.31 74.8 35.60 2.32 3.24 0.0531 0.1404
B18 126.67 78.5 43.02 2.60 3.62 0.0582 0.4514
B19 120.00 77.2 42.76 2.71 3.66 0.0627 0.1149
B20 71.43 75.8 35.80 1.84 3.20 0.0617 0.2811
B21 100.00 77.7 41.57 2.04 3.38 0.0672 0.1512
Median 100.00 76.8 38.08 2.11 3.24 0.0628 0.1404
IQR 12.24 1.73 3.46 0.22 0.13 0.0106 0.2016

response values, and the independent variables, psycho- where MA is the discomfort value, L is the loudness, S is the
acoustic metrics, for all noise stimuli in both aeroplanes. sharpness, and R is the roughness for both planes.
The variables of the MA model are determined by consider- The results of the t-test, the VIF, the F-test and multiple
ing the partial correlation coefficient, F-test and t-test coefficient of determination (i.e., R2, calculated by the linear
results, and VIF. least square method for the difference between the measured
The MA model for total data is shown as follows: and predicted dependent values) for the variables and equa-
tion are shown in Table V. The high determination coeffi-
MA ¼ 68:010 þ 3:658L þ 19:842S  47:426R; (9) cient (R2), and the significant F-test results illustrate that the

FIG. 7. (Color online) The heatmaps of the median response values (above) and A-weighted sound pressure levels (below) for each measuring point in the
aircraft cabin. The air conditioning outlet points are marked with the symbol .

1970 J. Acoust. Soc. Am. 154 (4), October 2023 Huang et al.
https://doi.org/10.1121/10.0020838

TABLE III. Correlation matrix between discomfort response value, A-weighted sound pressure level [SPL(A)], and psychoacoustic metrics, i.e., loudness
(L), sharpness (S), roughness (R), fluctuation strength (FS), and tonality for planes A and B. *p < 0.05, **p < 0.01.

Plane A Discomfort SPL(A) L S R FS Tonality

Discomfort 1 — — — — — —
SPL(A) 0.642** 1 — — — — —
L 0.827** 0.907** 1 — — — —
S 0.322 0.194 0.198 1 — — —
R 0.162 0.296 0.138 0.694** 1 — —
FS 0.766** 0.849** 0.848** 0.454* 0.171 1 —
Tonality 0.713** 0.294 0.599** 0.532** 0.589** 0.583** 1

Plane B Discomfort SPL(A) L S R FS Tonality

Discomfort 1 — — — — —
SPL(A) 0.489** 1
L 0.678** 0.885** 1 — — — —
S 0.267 0.003 0.217 1 — — —
R 0.085 0.547* 0.591* 0.588* 1 — —
FS 0.528* 0.621** 0.512* 0.364 0.077 1 —
Tonality 0.387 0.540* 0.503* 0.029 0.381 0.571* 1

MLR model well estimated the acoustic discomfort in the models. The mean squared error (MSE) and correlation
aircraft cabin, and all independent variables significantly coefficient (r) of each model are also marked in Fig. 8. The
influence the response values (p < 0.05, t-test). The small linear MA model performs better with higher correlation
VIF values indicate no multicollinearity of the equation coefficients and lower MSE than the nonlinear PA models.
obtained from MLR approaches. Noting that the indepen- Inspired by More approaches,22 we modified these PA
dent variable sharpness had a significant (p ¼ 0.03, t-test, as models by introducing the coefficients for independent vari-
shown in Table V) effect on discomfort and would increase ables. The modified PA models with various c0 are as
the determination coefficient of the model during the step- following:
wise multiple regression approach, thus we adopted sharp- The modified Zwicker PA model,
ness as an independent variable despite its relatively low  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
partial correlation coefficient with discomfort (0.116 when PAmod ¼ N5 1 þ c0 0 þ c1 0 xS 2 þ c2 0 xFR 2 ; (10)
controlling for loudness, as shown in Table IV).
We also estimated response values using the Zwicker the modified More PA model,
PA model, More PA model, and Di PA model in the forms  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
of Eqs. (1), (4), and (6), respectively. Figure 8 demonstrates PAMmod ¼ N5 1 þ c0 0 þ c1 0 xS 2 þ c2 0 xFR 2 þ c3 0 xTM 2 ;
the normalised measured and estimated response values
(11)
obtained from the experiment and different sound quality
h i
ð c4 0 N5 Þ2 ð c5 0 K5 Þ2
TABLE IV. Zero-order (controlling for none variable) and partial correla- xTM 2 ¼ 1e 1e ; (12)
tion coefficients (controlling for loudness) between discomfort response
values and psychoacoustic metrics, i.e., sharpness, roughness, FS, tonality and the modified Di PA model,
for planes A and B. *p < 0.05, **p < 0.01.
 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Plane A Loudness Sharpness Roughness FS Tonality PADmod ¼ N5 1 þ c0 0 þ c1 0 xS 2 þ c2 0 xFR 2 þ c3 0 xTD 2 ;
Zero-order 0.827** 0.322 0.162 0.766** 0.713** (13)
Controlling — 0.286 0.498* 0.216 0.484*
for loudness
TABLE V. The values of the variables, t-test, VIF, F-test, and determina-
Plane B Loudness Sharpness Roughness FS Tonality
tion coefficient (R2) of the sound quality models for acoustic discomfort in
Zero-order 0.678** 0.267 0.085 0.528** 0.387 the aircraft cabin.
Controlling — 0.167 0.546* 0.287 0.072
Model Variable t-test p-value VIF F-test p-value R2
for loudness
Total Intercept 2.137 0.040
Total Loudness Sharpness Roughness FS Tonality
Loudness 8.962 0.000 1.411
Zero-order 0.808** 0.375 0.263 0.713** 0.656** Sharpness 2.205 0.034 3.143
Controlling — 0.116 0.398* 0.275 0.298 Roughness 3.398 0.002 2.673
for loudness Eq. (9) MA ¼ 68:010 þ 3:658L þ 19:842S  47:426R 33.098 0.000 0.745

J. Acoust. Soc. Am. 154 (4), October 2023 Huang et al. 1971
https://doi.org/10.1121/10.0020838

FIG. 8. (Color online) Measured response values (i.e., inter-individual median value per stimuli) versus estimated values of discomfort: (a) MA model, (b)
Zwicker PA model (Ref. 10), (c) More PA model (Ref. 22), and (d) Di PA model (Ref. 23). Note that the data are normalized to a 0–1 interval.

where xS , xFR , and xTM , and the ci 0 (i ¼ 1–5) are the coeffi- expectation because people tend to prefer a steady noise to a
cient factors. The c0 coefficients in Eqs. (10)–(13) was deter- fast fluctuating one, e.g., a study on road traffic noise annoy-
mined using a nonlinear regression analysis with the ance that manipulated the stimuli to the same SPL artifi-
measured response values as the dependent variable and the cially found a positive correlation between roughness (R5
different SQMs [i.e., the x factors in Eqs. (10)–(13)] as value in the range 1.4–2.2 asper) and annoyance,43 similar
independent variables. The calculated values of the ci 0 are to another study on road traffic noise induced discomfort
shown in Table VI. The normalised measured and estimated and roughness (R50 value in the range 1.4–2.6 asper).46
response values, as well as the MSE and correlation coeffi- However, some previous studies also observed a similar
cient (r), are shown in Fig. 9. phenomenon as our present study, e.g., negative correlation
between roughness and noise discomfort on micro commer-
IV. DISCUSSION cial vehicles on the bump road,19 the acoustic pleasant posi-
tively correlated to roughness in the range 0.26–6.95 asper
A. The discomfort and psychoacoustic metrics of vehicle HVAC systems,21 and noise annoyance nega-
All the study participants proactively reported annoyed tively correlated to roughness in the multiple regression
or uncomfortable feelings caused by the stimuli after the sound quality model (a positive correlation found without
experiment, so we were convinced that they rated the dis- controlling loudness) for subway cars.47 A systematic study
comfort but not loudness. Tables III and IV show that dis- on aircraft noise characteristics conducted by More22
comfort response values strongly correlate to loudness in adjusted the roughness (in the range of 1.48–3.77 asper),
aircraft cabins. Previous studies have shown that annoyance while keeping loudness the same by applying frequency and
(discomfort) is significantly related to noise levels (e.g., amplitude modulations to time history and intensifying the
SPL and loudness) from laboratory experiments, in situ sur- fast fluctuations (50–90 per second) for the signal, and found
that the subjects fall into three groups according to subject-
veys, and meta-analyses.17,26,38–41 Simple linear regressions
to-roughness correlation coefficients, i.e., 21 out of 37 par-
were used further to investigate the influence of each SQM
ticipants were more annoyed with rougher sounds (i.e., posi-
on discomfort. Figure 10 shows that both for Plane A and
tive correlation between annoyance and roughness), nine
Plane B, the R2 of simple linear regression analysis between
participants were the opposite (i.e., negative correlation
response values and loudness is the greatest among that
between annoyance and roughness), and seven participants
between response values and each independent variable
appeared to be not sensitive to changes in roughness. There
(i.e., SPL, loudness, sharpness, roughness, fluctuation
seems to be an underlying psychophysical mechanism other
strength, and tonality). It is consistent with studies on air-
than mathematical phenomena such as multicollinearity.
craft interior noise,42 and rotor noise,24 indicating loudness A recent study with a series of 1-s click trains found
dominates the discomfort response values. that the relationship between the roughness and subjective
From Eq. (9) and Table IV (when controlling loudness), aversion is nonmonotonic in the frequency range from
the discomfort response is negatively correlated to rough- 10 to 140 Hz, indicating that the aversion decreases with
ness in the 2.52–3.66 asper range. This finding is out of increasing amplitude modulations faster than 40 Hz (in the
roughness range from 30 to 140 Hz).44 It merits thinking
TABLE VI. The c coefficients calculated by nonlinear least square method. whether the discomfort, which is also a quantification merit
of negative emotion, would have a similar relationship with
Model c0 0 c1 0 c2 0 c3 0 c4 0 c5 0 the amplitude modulation frequency. The roughness value
Modified Zwicker PA 6.671 18.427 4.927 — — — (R) is proportional to the frequency of modulation (fmod ) and
Modified More PA 8.805 17.433 5.697 1.580 10.000 8.045 temporal masking depth (DL)10,
Modified Di PA 10.010 24.333 8.351 0.124 — —
R  fmod DL: (14)

1972 J. Acoust. Soc. Am. 154 (4), October 2023 Huang et al.
https://doi.org/10.1121/10.0020838

FIG. 9. (Color online) Measured response values (i.e., inter-individual median value per stimuli) versus estimated values of discomfort: (a) MA model, (b)
modified Zwicker PA model, (c) modified More PA model, and (d) modified Di PA model. Note that the data are normalized to a 0–1 interval.

The frequency component (the carrier frequency or the mod- B. The sound quality models for aircraft cabin noise
ulation frequency) of the aircraft noise that contributes discomfort
greatly to the roughness does not necessarily produce great Comprehensively considering the significance of the t-test
discomfort, e.g., the components around 1080–2320 Hz (the for each variable, the F-test for the regression equation, the
10–14th bark) for stimuli A02 and B03, which are not the VIF, and the determination coefficient R2 in the linear regres-
primary energy components for the aircraft cabin noise. sion analysis, we selected loudness, sharpness, and roughness
Different modulation frequencies may lead to other rough or as the independent variables of the MA model. This MA model
aversion behaviours and activate responses in different brain performed better in discomfort estimation (R2 ¼ 0.745, Table
regions, e.g., the study by Arnal et al.,44 resulting in a com- V) than a simple linear regression model with an independent
plex neural mechanism. The same roughness values with variable of loudness (with R2 ¼ 0.657, Fig. 10), and would not
different carrier and modulation frequencies may not neces- cause multicollinearity (VIF < 5, Table V). It might be argu-
sarily lead to identical discomfort responses, which might able whether participants could perceive such a small variation
be a reason for the significant effect of the relatively small range of roughness or sharpness (see the IQRs in Tables I and
range of roughness value (i.e., 2.52–3.66 asper) on discom- II) during the tests. Although it is inconclusive that the small
fort responses in this study. Thus, investigating by strictly variations of roughness and sharpness affect the acoustic dis-
controlling the noise characteristics (i.e., SPL and loudness) comfort in reality, at least the present MA model with three
is necessary to obtain further evidence on the effects of independent variables is practical and might bring benefits in
roughness on perceived acoustic discomfort. distinguishing discomfort of noise having similar loudness.

FIG. 10. (Color online) The diagrams and R2 of simple linear regression analyses between values of discomfort and metrics of sound pressure level, loud-
ness, sharpness, roughness, fluctuation strength, and tonality. Bronze solid line front section, olive solid line middle section, maroon solid line rear section,
䊉, sitting position near the window, 䉬, standing position in the aisle, w, air conditioning outlet.

J. Acoust. Soc. Am. 154 (4), October 2023 Huang et al. 1973
https://doi.org/10.1121/10.0020838

It might also be arguable that the perceived tonality TABLE VII. The values of the variables, t-test, VIF, F-test, and determina-
showed a large difference among stimuli at different posi- tion coefficient (R2) of the MA linear regression models using the first half
data.
tions (Tables I and II), so it seemed to correlate significantly
with discomfort in plane A (0.713** in Table III) on dis- Model Variable t-test p-value VIF F-test p-value R2
comfort. However, we think such an association between
Total Intercept 0.686 0.503
tonality and discomfort response is largely confounded by
Loudness 6.870 0.000 1.759
loudness. When controlling for loudness, the significant cor-
Sharpness 1.625 0.125 3.404
relation between tonality and discomfort of all stimuli disap- Roughness 2.202 0.044 2.888
peared (correlation coefficients reduced from 0.696** to Eq. (A1) MA ¼ 33:346 þ 4:764L þ 22:987S  51:401R 20.693 0.000 0.805
0.258, as shown in Table III). During the stepwise multiple
linear regression approach, adding the tonality as an inde-
pendent variable neither increased the determination coeffi- metrics if not considering the multicollinearity, e.g., R2 would
cient (R2) nor significantly influenced discomfort (p ¼ 0.45, be 0.741 when adding the tonality in the MLR procedure.
t-test). Therefore, we excluded tonality as a good indicator All the MA and modified PA models reasonably esti-
of aircraft cabin noise. mate the noise discomfort in the aircraft cabin. The MA
We illustrated the validity and robustness of the MA model is intuitive in showing the contribution of parameters
and modified PA models this study established from two and evaluating the noise discomfort for a specific object.
aspects. First, we used the data of the air conditioning outlet The PA model is also applicable and could be modified by
points (i.e., index B18–B21 in Table II), which were not adding new parameters and data to achieve better perfor-
adopted in the modelling. The correlation coefficients, r, mance in estimating noise discomfort without considering
between the estimated and measured values were 0.968, the multicollinearity problem. However, existing PA models
0.968, 0.930, and 0.965 for the MA model, modified can hardly illustrate the relationship between a specific
Zwicker PA model, modified More PA model, and modified SQM and discomfort, e.g., the negative correlation between
Di PA model, respectively. Second, we split the data in half: roughness and discomfort caused by the aircraft cabin noise.
A01–07, A15, A18, A20, B01–06, B13, B14, and B16 (i.e., The modified PA models with coefficients for independent
the first half) versus data of A08–14, A16, A17, A19, A21, variables [i.e., Eqs. (10)–(13) and Table VI] may overcome
B07–12, B15, and B17 (i.e., the second half), as indexed in this shortage.
Fig. 2, and found good Cronbach’s alpha (i.e., 0.82) between
the two halves of the data. We fit the MA and modified PA C. The noise discomfort in different positions
models using the first half data and validated the models of the aircraft cabin
using the second half data, as shown in the Appendix. The
In Secs. III A and III B, the noise caused the most dis-
correlation coefficients, r, between the estimated and mea-
comfort responses in the middle cabin section, and discom-
sured values of the validation sets were 0.871, 0.887, 0.908,
fort at the window position was greater than at the aisle
and 0.890 for the MA model, modified Zwicker PA model,
position for the usual activity positions of passengers, indi-
modified More PA model, and modified Di PA model,
cating a priority target for acoustic comfort design.
respectively.
Moreover, it seems no longer appropriate to adopt the A-
As listed in Table VI, the MA model better estimates
weighted SPL purely as the reference indicator, e.g., the
acoustic discomfort response with greater correlation coeffi-
noise in plane B (except for the air conditioning outlet posi-
cients and less MSE than the PA models. A possible reason
tions) had a slight variation of A-weighted SPL (min to
is that these PA models were not established for aircraft
mum, less than 3 dBA) but a big difference in discomfort, as
cabin noise, e.g., More PA model was established by inves-
shown in Table II and Fig. 7. Even the opposite was found,
tigating aircraft traffic noise, and Di PA model was estab-
e.g., A02 had a slightly lower A-weighted SPL [i.e., around
lished from transformer radiated noise. The modified
0.7 dB(A) less] but much higher discomfort value (i.e.,
models performed better in discomfort estimation with 20%
around 37.9% points higher) than A04 in plane A, and B11
points higher correlation coefficients and lower MSEs com-
had the greatest A-weighted SPL but relatively small dis-
pared to the original PA models after introducing the c0
comfort among all the noise stimuli in plane B, as shown in
coefficients. The subjective evaluation is essential in the
Tables II and III and Fig. 7. At least, if not the discomfort
sound quality modelling of noise discomfort since the c0 val-
values, the loudness is essential to guide the acoustic design.
ues were iterated by the nonlinear least square method using
the reported values from the experiment. From Fig. 9 and
TABLE VIII. The c coefficients in nonlinear regression models using the
Table VIII, the modified PA models perform slightly better first half data.
than the MA model in estimating the acoustic discomfort in
the aircraft cabin, and the More and Di models perform Model c0 0 c1 0 c2 0 c3 0 c4 0 c5 0
slightly better than the Zwicker PA model by introducing
Modified Zwicker PA 8.312 20.098 6.493 — — —
tonality. Higher R2 and lower MSE are expected to benefit Modified More PA 10.163 19.972 7.664 0.934 10.000 16.082
from adopting more SQMs as independent variables. We Modified Di PA 8.718 20.548 6.901 0.019 — —
could increase the R2 of the MA model by adopting more
1974 J. Acoust. Soc. Am. 154 (4), October 2023 Huang et al.
https://doi.org/10.1121/10.0020838

FIG. 11. (Color online) Measured response values (i.e., inter-individual median value per stimuli) versus estimated values of the second half data. The dis-
comfort models were established using the first half data: (a) MA model, (b) modified Zwicker PA model, (c) modified More PA model, (d) modified Di PA
model. Note that the data are normalized to a 0–1 interval.

The greatest discomfort response values at the air con- B13, B14, and B16 (i.e., the first half) and validated using data
ditioning outlet positions, as shown in Sec. III A, indicated of A08–14, A16, A17, A19, A21, B07–12, B15, and B17 (i.e.,
the major contribution of the environmental control system the second half), as indexed in Fig. 2. The normalised mea-
to acoustic discomfort in the aircraft cabin,25 which merited sured and estimated response values, as well as the MSE and
the primary concern. In this study, the discomfort caused by correlation coefficient (r), are shown in Fig. 11.
noise in the front section is significantly smaller than in the
1
middle and rear sections. However, a previous study found Airbus (2023). “Cabin comfort can make such a difference to the passen-
ger experience. This is why it is a major priority for Airbus,” https://
that noise in the rear section was less unpleasant (more www.airbus.com/en/products-services/commercial-aircraft/cabin-and-co
acceptable) than in the front or middle section.26 It seems mfort (Last viewed August 31, 2023.).
2
that the acoustic discomfort response varied much, with its J. Bouwens, S. Hiemstra-van Mastrigt, and P. Vink, “Ranking of human
characteristics depending on the aircraft. It merits more spe- senses in relation to different in-flight activities contributing to the com-
fort experience of airplane passengers,” Int. J. Aviat. Aeronaut. Aerosp.
cific investigation on different aircraft types in further study. 5(2), 9 (2018).
3
P. Vink, C. Bazley, I. Kamp, and M. Blok, “Possibilities to improve the
V. CONCLUSION aircraft interior comfort experience,” Appl. Ergon. 43(2), 354–359
(2012).
The sound quality models for noise-induced discomfort in 4
V. Mellert, I. Baumann, N. Freese, and R. Weber, “Impact of sound and
the cabins of two large passenger aeroplanes were established vibration on health, travel comfort and performance of flight attendants
and compared using MLR and various PA approaches, respec- and pilots,” Aerosp. Sci. Technol. 12(1), 18–25 (2008).
5
J. Quehl, Comfort Studies on Aircraft Interior Sound and Vibration
tively. The MA model better estimated the subjective response (Universit€at Oldenburg, Oldenburg, Germany, 2001).
than the Zwicker, More, and Di PA models. After introducing 6
D. Berckmans, K. Janssens, A. H. Van, P. Sas, and W. Desmet, “Model-
appropriate coefficients for independent variables, the perfor- based synthesis of aircraft noise to quantify human perception of sound
quality and annoyance,” J. Sound Vib. 311(3–5), 1175–1195 (2008).
mance of all PA models improved. The acoustic discomfort in 7
K. White, A. W. Bronkhorst, and M. Meeter, “Annoyance by transporta-
the aircraft cabin was dominated by loudness and was influ- tion noise: The effects of source identity and tonal components,”
enced by sharpness positively and by roughness negatively. J. Acoust. Soc. Am. 141(5), 3137–3144 (2017).
8
Reducing loudness and sharpness (i.e., the proportion of high- A. J. Torija, S. Roberts, R. Woodward, I. H. Flindell, A. R. McKenzie,
and R. H. Self, “On the assessment of subjective response to tonal content
frequency sound energy components) may improve acoustic
of contemporary aircraft noise,” Appl. Acoust. 146, 190–203 (2019).
comfort better than merely controlling the A-weighted SPL, 9
Y. Huang, G. Di, Y. Zhu, Y. Hong, and B. J. Zhang, “Pair-wise compari-
whereas increasing or reducing roughness may be cautiously son experiment on subjective annoyance rating of noise samples with dif-
approached for the aircraft cabin noise. The negative correla- ferent frequency spectrums but same A-weighted level,” Appl. Acoust.
69(12), 1205–1211 (2008).
tion between discomfort and roughness merits careful investi- 10
Y. Soeta and H. Kagawa, “Three dimensional psychological evaluation of
gation in further study. aircraft noise and prediction by physical parameters,” Build. Environ.
167, 106445 (2020).
11
ACKNOWLEDGMENTS M. Brink, B. Sch€affer, D. Vienneau, M. Foraster, R. Pieren, I. C. Eze, C.
Cajochen, N. Probst-Hensch, M. R€ o€
osli, and J. M. Wunderli, “A survey
We gratefully acknowledge the National Natural on exposure-response relationships for road, rail, and aircraft noise annoy-
Science Foundation of China Grant No. 52072242 and A ance: Differences between continuous and intermittent noise,” Environ.
Int. 125, 277–290 (2019).
Ministry Key Project (GW0890006). 12
E. Zwicker and H. Fastl, Psychoacoustics: Facts and Models (Springer
Science & Business Media, Heidelberg, Germany, 2013).
13
S. C. Lee, J. Y. Hong, and J. Y. Jeon, “Effects of acoustic characteristics
APPENDIX of combined construction noise on annoyance,” Build. Environ. 92,
657–667 (2015).
The MA linear regression model and modified PA nonlin- 14
Y. Soeta and R. Shimokura, “Sound quality evaluation of air-conditioner
ear regression models, as shown in Tables VII and VIII, were noise based on factors of the autocorrelation function,” Appl. Acoust.
established using data of A01–07, A15, A18, A20, B01–06, 124, 11–19 (2017).

J. Acoust. Soc. Am. 154 (4), October 2023 Huang et al. 1975
https://doi.org/10.1121/10.0020838

15 32
S. Sato, J. You, and J. Jeon, “Sound quality characteristics of refrigerator ECMA-418-2: Psychoacoustic Metrics for ITT Equipment — Part 2
noise in real living environments with relation to psychoacoustical and (Models Based on Human Perception) (Ecma International, Geneva,
autocorrelation function parameters,” J. Acoust. Soc. Am. 122, 314–325 Switzerland, 2020).
33
(2007). R. Sottek, Modelle Zur Signalverarbeitung im Menschlichen Geh€ or
16
H. Li, K. Chen, X. Wang, Y. Gao, and W. Yu, “A perceptual dissimilar- (Models for Signal Processing in Human Hearing) (RWTH Aachen,
ities based nonlinear sound quality model for range hood noise,” Aachen, Germany, 1993).
34
J. Acoust. Soc. Am. 144(4), 2300–2311 (2018). Y. Huang and M. J. Griffin, “The relative discomfort of noise and vibra-
17
Y. Huang and Q. Zheng, “Sound quality modelling of hairdryer noise,” tion: Effects of stimulus duration,” Ergonomics 57(8), 1244–1255 (2014).
35
Appl. Acoust. 197, 108904 (2022). A. Sen and M. Srivastava, Regression Analysis: Theory, Methods, and
18
Z. M. Nopiah, A. K. Junoh, and A. K. Ariffin, “Noise annoyance fuzzy Applications (Springer Science & Business Media, Berlin, 1997).
36
index in passenger car cabin,” Int. J. Veh. Noise Vib. 9(3–4), 216–233 S. S. Castellan, Nonparametric Statistics for the Social Sciences
(2013). (McGraw-Hill, New York, 1988).
19 37
D. Li and Y. Huang, “The discomfort model of the micro commercial H. Hofmann, H. Wickham, and K. Kafadar, “Letter-value plots: Boxplots
vehicles interior noise based on the sound quality analyses,” Appl. for large data,” J. Comput. Graph. Stat. 26, 469–477 (2017).
38
Acoust. 132, 223–231 (2018). E. Maris, P. J. Stallen, R. Vermunt, and H. Steensma, “Noise within the
20
E. Zhang, Q. Zhang, J. Xiao, L. Hou, and T. Guo, “Acoustic comfort eval- social context: Annoyance reduction through fair procedures,” J. Acoust.
uation modeling and improvement test of a forklift based on rank score Soc. Am 121(4), 2000–2010 (2007).
39
comparison and multiple linear regression,” Appl. Acoust. 135, 29–36 J. K. Lee, J. Jang, S. I. Chang, and S. I. Lee, “Annoyance modeling using
(2018). personal and situational variables for construction site noise in urban
21
J. H. Yoon, I. H. Yang, J. E. Jeong, S. G. Park, and J. E. Oh, “Reliability areas,” Appl. Acoust. 182, 108256 (2021).
40
improvement of a sound quality index for a vehicle HVAC system using a Y. Wang, S. Zhang, D. Meng, and L. Zhang, “Nonlinear overall annoy-
regression and neural network model,” Appl. Acoust. 73(11), 1099–1103 ance level modeling and interior sound quality prediction for pure electric
(2012). vehicle with extreme gradient boosting algorithm,” Appl. Acoust 195,
22
S. R. More, Aircraft Noise Characteristics and Metrics (Purdue 108857 (2022).
41
University, West Lafayette, IN, 2011). R. Guski, D. Schreckenberg, and R. Schuemer, “WHO environmental
23
G. Q. Di, X. W. Chen, K. Song, B. Zhou, and C. M. Pei, “Improvement of noise guidelines for the European Region: A systematic review on envi-
Zwicker’s psychoacoustic annoyance model aiming at tonal noises,” ronmental noise and annoyance,” Int. J. Environ. Res. Public Health
Appl. Acoust. 105, 164–170 (2016). 14(12), 1539 (2017).
24 42
A. J. Torija, Z. Li, and P. Chaitanya, “Psychoacoustic modelling of rotor J. R. Angerer, D. A. Mccurdy, and R. A. Erickson, “Development of an
noise,” J. Acoust. Soc. Am. 151(3), 1804–1815 (2022). annoyance model based upon elementary auditory sensations for steady-
25
B. O. de Sa Lopes, Sound Quality Predictive Models: Annoyance in the state aircraft interior noise containing tonal components,” NASA TM-
Interior of a Propeller Aircraft (University of Porto, Porto, Portugal, 104147 (NASA, Washington, DC, 1991).
43
2018). J. Felcyn, “The influence of a signal’s time structure on the perceived
26
S. Pennig, J. Quehl, and V. Rolny, “Effects of aircraft cabin noise on pas- noise annoyance of road traffic noise,” J. Environ. Health Sci. Eng. 19,
senger comfort,” Ergonomics 55(10), 1252–1265 (2012). 881–892 (2021).
27 44
S. S. Stevens, Psychophysics: Introduction to its Perceptual, Neural, And L. H. Arnal, A. Kleinschmidt, L. Spinelli, A.-L. Giraud, and P.
Social Prospects (Wiley, New York, 1986). Megevand, “The rough sound of salience enhances aversion through neu-
28
Y. Huang and M. J. Griffin, “Comparison of absolute magnitude estima- ral synchronization,” Nat. Commun. 10, 3697 (2019).
45
tion and relative magnitude estimation for judging the subjective intensity A. J. Torija, P. Chaitanya, and Z. Li, “Psychoacoustic analysis of contra-
of noise and vibration,” Appl. Acoust. 77, 82–88 (2014). rotating propeller noise for unmanned aerial vehicles,” J. Acoust. Soc.
29
Y. Huang and D. Li, “An empirical category-ratio scale for evaluating the Am. 149(2), 835–846 (2021).
46
subjective intensity of noise based on the comparison of estimated magni- R. B. Raggam, M. Cik, R. R. H€ oldrich, K. Fallast, E. Gallasch, M. Fend,
tudes and categories,” Appl. Acoust. 158, 107048 (2020). A. Lackner, and E. Marth, “Personal noise ranking of road traffic:
30
ISO 532-1-2017: Acoustics—Methods for calculating loudness—Part 1: Subjective estimation versus physiological parameters under laboratory
Zwicker method (ISO, Geneva, Switzerland, 2017). conditions,” Int. J. Hygiene Environ. Health 210(2), 97–105 (2007).
31 47
DIN 45692: Measurement Technique for the Simulation of the Auditory L. K. Rimskaya-Korsakova, P. A. Pyatakov, and S. A. Shulyapov,
Sensation of Sharpness (Deutsches Institut F€ ur Normung, Berlin, “Evaluations of the annoyance effects of noise,” Acoust. Phys. 68,
Germany, 2009). 502–512 (2022).

1976 J. Acoust. Soc. Am. 154 (4), October 2023 Huang et al.

You might also like