HAL Id: hal-03234182
https://hal.archives-ouvertes.fr/hal-03234182
Submitted on 26 May 2021HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci-entific research documents, whether they are pub-lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Multidimensional psychological evaluation of
air-conditioner sounds
Yoshiharu Soeta, Ei Onogawa
To cite this version:
MULTIDIMENSIONAL PSYCHOLOGICAL EVALUATION OF
AIR-CONDITIONER SOUNDS
Yoshiharu soeta
1Ei Onogawa
21 National Institute of Advanced Industrial Science and Technology (AIST), Osaka, Japan 2 Mitsubishi Heavy Industries, Nagoya, Japan
Correspondence .soeta@aist.go.jp
ABSTRACT
The aim of this paper is to clarify the characteristics of air-conditioner sounds and determine the factor that is most influential on the subjective response caused by the sound. The A-weighted equivalent continuous sound pressure level (LAeq) and factors extracted from the autocorrelation
function (ACF) and interaural cross-correlation (IACF) were analyzed. Subjective loudness, clearness, continuity, pitch, sharpness, and annoyance were evaluated using a paired comparison method. Multiple regression analyses were performed using a linear combination of LAeq, the
ACF factors, IACF factors, and their standard deviations. The results indicated that the acoustic characteristics caused by air conditioners were characterized by the ACF and IACF factors. The multiple regression analyses indicated that the subjective responses caused by air-conditioner sounds can be predicted by LAeq, the delay time
and amplitude of the first maximum peak, the width of the first decay of the ACF, and the magnitude of the IACF.
1. INTRODUCTION
Large electrical appliances such as air conditioners, refrigerators, and washing machines are regarded as major noise sources in building environments. Air conditioners are widely used in offices and residences for long periods, and thus considerable efforts have been made to reduce the sound levels of these devices during operation. As results, the sound levels of air conditioners are now comparatively low [1, 2]. However, people may still be made to feel uncomfortable by certain aspects of the sound quality, even when the actual sound level of the air conditioners is low [3]. Therefore, both the sound levels and the sound quality of an air conditioner are important for the user’s acoustic comfort.
Previous studies have evaluated the relationships between sound quality metrics (termed psychoacoustic factors) such as loudness, sharpness, and roughness and subjective responses for air-conditioner noises [4-6]. The results indicated that psychoacoustic factors can influence subjective responses. As with other psychoacoustic factors, autocorrelation function (ACF) factors are significantly correlated with subjective responses [7,8].
The aim of this study is to determine the ACF and interaural cross-correlation (IACF) factors that is the most dominant in terms of the subjective responses caused by air-conditioner sounds. One rationale for the ACF and IACF approach is that the factors describe the basic temporal sensations, such as loudness and pitch [9, 10].
2. METHODS 2.1 Analysis of air-conditioner sounds
Air-conditioner sounds generated by three outlet units and one inlet unit were measured at different conditions. e.g., revolutions per minute of the fan and compressor, using a binaural microphone (BHS I, HEAD Acoustics). For all measurements, the noise was recorded via an analog-to-digital/digital-to-analog (AD/DA) converter (SQuadrigaII, HEAD Acoustics) at a sampling rate of 48 kHz and with a sampling resolution of 32 bits.
Factors extracted from ACF and IACF have been proposed for sound quality evaluation [9, 10]. To calculate the ACF and IACF factors, the normalized ACF and IACF of the signals recorded at left and right ears from the microphones, pl(t) and pr(t), as a function of the running
step, s, is defined by Ilr(W) = Ilr(W; s, T) = ఃೝሺఛǢ௦ǡ்ሻ ඥఃሺǢ௦ǡ்ሻఃೝೝሺǢ௦ାఛǡ்ሻ, (1) where ߔሺ߬Ǣ ݏǡ ܶሻ ൌ ଵ ଶ் ′ሺݐሻ ′ሺݐ ߬ሻ ௦ା் ௦ି் ݀ݐ. (2)
When only the signal recorded at the left, pl(t), or right,
pr(t), ear, is used, the equation (1) shows the normalized
ACF. Here, 2T is the integration interval and p’(t) = p(t)*se(t), where se(t) is the ear sensitivity. pl(t) and pr(t) is
the signal that was measured using the binaural microphone in this study. se(t) represents the impulse
response of an A-weighted network, including the transfer functions of the human outer and middle ear, for convenience [9, 10]. Then normalized ACF and IACF are carried out using the geometric mean of the energy at s and the energy at s+W. This ensures that the normalized ACF and IACF satisfy the condition 0 ≤ I(τ) ≤ 1.
LAeq was determined from the A-weighted p(t) as a
function of s. LAeq is calculated using
LAeq(s, T) = ͳͲඥߔሺሻሺͲǢ ݏǡ ܶሻ. (3)
This means that the ACF includes LAeq as one of its factors.
The other ACF factors are calculated from the normalized ACF as shown in Figure 1 (a). W1 and I1 are defined as the
delay time and the amplitude of the first maximum peak. W1 and I1 are related to the perceived pitch and the pitch
strength of sounds, respectively [9, 10]. Higher values of W1 and I1 mean that the sound has a lower or a stronger
delay and represents a repetitive feature containing the sound source itself [9]. The other ACF factor, the width of the first decay, WI(0), is defined using the delay time
interval at a normalized ACF value of 0.5. WI(0) is
equivalent to the spectral centroid [10]. Higher values of WI(0) indicate that the sound includes a higher proportion
of low-frequency components.
The interaural cross-correlation coefficient (IACC) is related to the subjective diffuseness and apparent source width [9], and is defined by
IACC(s, T) = |Ilr(W; s, T)|max, |W| ӊ 1 [ms]. (4)
When the IACC is 1, a listener can clearly perceive the direction of the sound source. When IACC approaches 0, listeners can hear the sound, but it is diffused. The other IACF factors are calculated from the normalized IACF as shown in Figure 1 (b). The interaural delay time at which IACC is defined is τIACC, corresponding to the sense of
direction at low frequency [9]. WIACC is the width of the
IACF defined by the interval of delay time at a value of δ below the IACC that may correspond to the just noticeable difference of the IACC. WIACC mainly depends on the
frequency component of the signals and is related to the apparent source width [9].
We calculated LAeq, W1, I1, WI(0), We, IACC, τIACC, and
WIACC as a function of time to evaluate the noise both
quantitatively and qualitatively. The integration interval, 2T, was 0.5 s and the running step, s, was 0.1 s in all calculations. The analyses were conducted using a Matlab-based analysis program (Mathworks, Natick, MA).
2.2 Subjective assessments
Fifteen stimuli were selected from the measured air-conditioner noise based on the distribution of the ACF and
IACF factors. The stimuli were presented binaurally through a headphone amplifier (HDVD800, Sennheiser, Germany) and headphones (HD800, Sennheiser, Germany). The duration of each stimulus was 2.0 s, including rise and fall ramps of 0.1 s. The participants sat in a comfortable thermal environment in a soundproof room to listen to the stimuli. All stimuli were presented at the same LAeq r 0.2 dB as the actual measured noises. LAeq
was verified using a dummy head microphone (KU100, Neumann, Germany) and a sound calibrator (Type 4231, Brüel & Kjær, Denmark).
We selected loudness, clearness, continuity, pitch, sharpness, and annoyance as the perceptual dimensions because they are used in daily evaluations. Subjective loudness, clearness, continuity, pitch, sharpness, and annoyance caused by air-conditioner sounds were evaluated to clarify the effects of the ACF and IACF factors on each subjective response. Participants with normal hearing, no history of neurological diseases, and an age range of between 20 and 54 years took part in the experiments. Fifteen participants took part in the loudness, clearness, sharpness, and annoyance experiment; and sixteen in the continuity and pitch experiment. Informed consent was obtained from each participant after the nature of the study was explained. The study was approved by the ethics committee of the National Institute of Advanced Industrial Science and Technology (AIST) of Japan.
Scheffe’s paired comparison tests [11] were carried out for all combinations of pairs (i.e., 105 pairs (N(N1)/2, N = 15)) of stimuli by interchanging the order in which the stimuli in each pair were presented and presenting the pairs in random order. The silent interval between stimuli was 1.0 s. Using a seven-point scale, the participants were asked to judge which stimulus from each pair was louder, clearer, rougher, higher, sharper, or more annoying, following presentation of each pair. The approximate duration of a single session was thirty minutes.
Scheffe's method [10] assigns one combination to each participant for comparison. In the modified Scheffe’s method, one participant performs a pairwise comparison of one iteration, and repeats this while changing the participant [11]. The averaged values of loudness, clearness, continuity, pitch, sharpness, and annoyance according to each participant were calculated based on the modified Scheffe’s method [10, 11] and were defined as the scale values (SVs). Analysis of variance (ANOVA) was then conducted on the results of the paired comparison experiments. The significance of the main effects of all possible combinations among stimuli was tested to check participant’s sensitivity to stimuli [11].
To calculate the effects of ACF and IACF factors on participant loudness, clearness, continuity, pitch, sharpness, and annoyance, multiple regression analyses were conducted using a linear combination of the mean ACF and IACF factors and their standard deviations (SDs) as predictive variables. To identify and quantify the significant objective factors of participant loudness, clearness, continuity, pitch, sharpness, and annoyance, stepwise selection was carried out by successively adding or removing variables. The stepping criteria employed for entry and removal were based on the significance level of Fig. 1. (a) The definition of the ACF factors, W1, I1, We,
and WI(0). (b) The definition of the IACF factors,
the F-value and set at 0.05 and 0.10, respectively. The analyses were carried out using SPSS statistical analysis software (SPSS version 22.0, IBM Corp., NY).
3. RESULTS
ANOVA for the scale values revealed that the main effect (i.e., the differences between the stimuli) was statistically significant (F(14, 2834) = 581.22, p < 0.001, for loudness, F(14, 2834) = 346.63, p < 0.001, for clearness, F(14, 3029) = 213.87, p < 0.001, for continuity, F(14, 3029) = 305.53, p < 0.001, for pitch, F(14, 2834) = 586.84, p < 0.001, for sharpness, F(14, 2834) = 390.80, p < 0.001, for annoyance). Figure 2 shows scale value of loudness, clearness, continuity, pitch, sharpness, and annoyance. Roughly speaking, loudness and annoyance showed similar tendency, clearness and sharpness showed similar tendency.
Figures 3, 4, and 5 show the relationship between each ACF/IACF factor and loudness, clearness, and continuity, respectively. Scale values of loudness had a high positive correlation with LAeq (r = 0.89, p < 0.01), W1
(r = 0.61, p < 0.01), and I1 (r = 0.59, p < 0.01). Scale values
of clearness had a high positive correlation with LAeq (r =
0.78, p < 0.01), W1 (r = 0.58, p < 0.01), and I1 (r = 0.63, p
< 0.01). Scale values of continuity had a high positive correlation with LAeq (r = 0.59, p < 0.01). Scale values of
pitch had a high positive correlation with LAeq (r = 0.57, p
< 0.01), I1 (r = 0.60, p < 0.01), and We (r = 0.54, p < 0.01).
Scale values of sharpness had a high positive correlation with LAeq (r = 0.73, p < 0.01), I1 (r = 0.69, p < 0.01), and
We (r = 0.59, p < 0.01). Scale values of annoyance had a
high positive correlation with LAeq (r = 0.82, p < 0.01) and
I1 (r = 0.52, p < 0.01).
A multiple linear regression analysis was performed with the SVs of loudness, clearness, continuity, pitch, sharpness, and annoyance for all participants as the outcome variable. The final models and the standardized partial regression coefficients were as follows:
SVloudness | a0 + 0.73LAeq 0.16SD_W1 + 0.24SD_IACC,
(5)
SVclearness | a1 + 0.61LAeq + 0.25W1 + 0.14IACC, (6)
SVcontinuity | a2 + 1.16LAeq 0.13W1 0.36I1 0.78We +
0.69SD_We, (7)
SVpitch | a3 + 0.55W1 + 0.34IACC + 0.19SD_W1 +
0.21SD_We, (8)
SVsharpness | a4 + 0.56LAeq + 0.26*We 0.14WI(0)
0.20SD_W1, (9)
SVannoyance | a5 + 0.75LAeq 0.10W1 + 0.34SD_WIACC
0.21SD_WI(0). (10)
The model was statistically significant (F(3, 221) = 410.62, p < 0.001, for loudness, (F(3, 221) = 135.44, p < 0.001, for clearness, F(5, 234) = 45.60, p < 0.001, for continuity, F(4, 235) = 66.66, p < 0.001, for pitch, F(4, 235) = 97.82, p < 0.001, for sharpness, F(4, 220) = 143.595, p < 0.001, p < 0.001, for annoyance), and the adjusted coefficient of determination, R2, was 0.85 for loudness, 0.64 for
clearness, 0.48 for continuity, 0.52 for pitch, 0.62 for sharpness, and 0.72 for annoyance.
Fig. 3. Relationship between ACF/IACF factors and scale value of loudness. (a) LAeq, (b) W1, (c) I1, (d) We, (e) WI(0),
(f) IACC, (g) WIACC, and (h) WIACC. Each symbol indicates mean value and error bars indicate SDs.
Fig. 4. Relationship between ACF/IACF factors and scale value of clearness. (a) LAeq, (b) W1, (c) I1, (d) We, (e) WI(0),
Fig. 5. Relationship between ACF/IACF factors and scale value of continuity. (a) LAeq, (b) W1, (c) I1, (d) We, (e) WI(0),
(f) IACC, (g) WIACC, and (h) WIACC. Each symbol indicates mean value and error bars indicate SDs.
Fig. 6. Relationship between ACF/IACF factors and scale value of pitch. (a) LAeq, (b) W1, (c) I1, (d) We, (e) WI(0), (f)
Fig. 7. Relationship between ACF/IACF factors and scale value of annoyance. (a) LAeq, (b) W1, (c) I1, (d) We, (e) WI(0),
(f) IACC, (g) WIACC, and (h) WIACC. Each symbol indicates mean value and error bars indicate SDs.
Fig. 8. Relationship between ACF/IACF factors and scale value of annoyance. (a) LAeq, (b) W1, (c) I1, (d) We, (e) WI(0),
4. DISCUSSION
A multiple linear regression analysis showed that the energy-index of LAeq is the significant factor for
multidimensional psychological evaluation of air-conditioner sounds except for pitch. The regression coefficients were all positive, suggesting that higher LAeq
causes clearer, rougher, sharper, and more annoying sound. Previous study indicated LAeq is not the significant factor
for annoyance [8], which is not consistent with the present finding. A possible reason for this could be due to different range of the LAeq, that is, the present study had higher and
broader range of LAeq values.
The ACF factor, W1, which is related to fundamental
frequency, is the significant factor for clearness, continuity, pitch, and annoyance prediction. The SD of W1 is the
significant factors for loudness, pitch, and sharpness prediction. These suggest not only the fundamental frequency, but also the temporal variation of the fundamental frequency has a large influence on subjective evaluation of air-conditioner sounds. The regression coefficient of W1 for annoyance was negative, which is
consistent with the previous finding [8].
The binaural index, IACC, is the significant factor for clearness and pitch, and the regression coefficient is positive. Sound with lower IACC values produces broader sound image [9]. Therefore, air-conditioner sounds with higher IACC values, which has narrower sound image, could cause clearer and higher pitch perception.
5. CONCLUSIONS
We analyzed psychological responses, loudness, clearness, continuity, pitch, sharpness, and annoyance of air-conditioner sounds to determine the factors that significantly influence subjective loudness, clearness, continuity, pitch, sharpness, and annoyance caused by this sounds. The results indicated that the LAeq, W1, and the
temporal variation of W1, and so on are factors that
significantly influence subjective responses. This indicates that the ACF and IACF factors are useful indices for the evaluation of air-conditioner sounds.
6. ACKNOWLEDGMENTS
This work was partly supported by a Grant-in-Aid for Scientific Research (B) (Grant No. 18H03324) from the Japan Society for the Promotion of Science.
7. REFERENCES
[1] U. Ayr, E. Cirillo, and F. Martellotta: “An experimental study on noise indices in air conditioned offices,” Appl Acoust, 62, 633-643, 2001.
[2] S.K. Tang, and M.Y. Wong: “On noise indices for domestic air conditioners,” J Sound Vib, 274, 1-12, 2004.
[3] T. Kitamura, R. Shimokura, S. Sato, and Y. Ando: “Measurement of temporal and spatial factors of a flushing toilet noise in a downstairs bedroom,” J Temp Des Arch Environ, 2, 13-19, 2002.
[4] E. Zwicker, and H. Fastl: Psychoacoustics: facts and models, Springer-Verlag, Berlin- Tokyo, (1999). [5] R.P. Leite, S. Paul, and S.N.Y. Gerges: “A sound
quality-based investigation of the HVAC system noise of an automobile model,” Appl Acoust, 645, 2009.
[6] J.H. Yoon, I.H. Yang, J.E. Jeong, S.G. Park, and J.E. Oh: “Reliability improvement of a sound quality index for a vehicle HVAC system using a regression and neural network model,” Appl Acoust 2012;73:1099-1103.
[7] S. Sato, J. You, and J.Y. Jeon: “Sound quality characteristics of refrigerator noise in real living environments with relation to psychoacoustical and autocorrelation function parameters,” J Acoust Soc Am, 122, 314-325, 2007.
[8] Y. Soeta, and R. Shimokura: “Sound quality evaluation of air-conditioner noise based on factors of the autocorrelation function,” Appl Acoust, 124, 11-19, 2017.
[9] Y. Ando, and P. Cariani: Auditory and visual sensations, Springer, New York, USA (2009). [10] Y. Soeta, and Y. Ando: Neurally based measurement
and evaluation of environmental noise, Springer, Tokyo, JPN (2015).
[11] H. Scheffé: “An analysis of variance for paired comparisons,” J Am Statist Assoc, 147, 381-400, 1952. [12] S. Sato: Statistical method of sensory test, Tokyo: