Challenge for the subjective experiment using headphone Susumu Hirakawa 1 Building Research Institute, Japan/ National Institute of Advanced Industrial Science and Technology (AIST) 1 Tachihara, Tsukuba, 3050802, Japan/ Building 2-1D, Central 2, 1-1-1 Umezono, Tsukuba, 3058568, Japan Hayato Sato 2 Environmental Acoustics Laboratory, Department of Architecture, Graduate School of Engineering, Kobe University Rokko, Nada, Kobe 657-8501, Japan Manabu Chikai 3 National Institute of Advanced Industrial Science and Technology (AIST) Building 6-11, Central 6, 1-1-1 Higashi, Tsukuba, 3058566, Japan Atsuo Hiramitsu 4 National Institute for Land and Infrastructure Management, 1, Tachihara, Tsukuba-City, Ibaraki, 305-0802, JAPAN Hiroshi Sato 5 National Institute of Advanced Industrial Science and Technology (AIST) Building 2-1D, Central 2, 1-1-1 Umezono, Tsukuba, 3058568, Japan Jeffrey Mahn 6 Markus Mueller-Trapet 7 Iara Batista da Cunha 8 Acoustics Group, Construction Research Centre, National Research Council Canada Building M27, 1200 Montreal Road Ottawa, Ontario K1A 0R6, Canada

1 Currently at National Institute for Land and Infrastructure Management. hirakawa-s92ta@mlit.go.jp

2 hayato@kobe-u.ac.jp

3 m-chikai@aist.go.jp

4 hiramitsu-a92ta@mlit.go.jp

5 sato.hiro@aist.go.jp

6 Jeffrey.Mahn@nrc-cnrc.gc.ca

7 Markus.Mueller-Trapet@nrc-cnrc.gc.ca

8 Iara.BatistadaCunha@nrc-cnrc.gc.ca

i, orn inter.noise 21-24 AUGUST SCOTTISH EVENT CAMPUS ? O? ? GLASGOW

ABSTRACT Due to the pandemic situation, it became complex to conduct subjective experiments in a laboratory setting which became inaccessible due to the lockdown. For this situation, the procedure for subjec- tive experiments without an anechoic chamber or listening room needs to be considered as an effec- tive alternative. A previous study in Canada showed that there is a good correlation between the online listening test on impact sound in the residential building using a headphone and the listening test conducted in the anechoic chamber using the loudspeaker. However, the study used the stimuli from an ambisonic microphone and not from a monaural microphone recording. Thus, further study was carried out with monaural and HATS recordings in Japan. The subjective evaluation in the anechoic chamber using a headphone was held in AIST, Tsukuba Japan. A total of 72 stimuli were created from 3 different floors, 2 different types of microphones (mono and HATS), at 12 different combinations of impact sources, excitation positions and micro- phone positions. The participants are 8 people. This study also provides some evidence and potential of the online/remote subjective experiments. The approach could be an effective alternative approach to the controlled laboratory experiment for impact sound. 1. INTRODUCTION

The listening test procedure requires a participant to be in a controlled environment and use reliable equipment. However, this requirement becomes a challenge due to Covid-19 restrictions. The listen- ing test setup in the laboratory also changed. The participant and researcher need to report the tem- perature for the past one or two weeks. The physical distance, ventilation and sanitization are also essential. Hence, the new listening test procedure which does not require the above requirement should be established for future preference. The preliminary challenge was conducted in Canada, which attempted to ensure the experience online (using headphones) gave similar results the experi- ence in the anechoic chamber [1]. The stimuli were the recording of the floor impact sound in the laboratory using a VR microphone. The signal was played in the anechoic chamber for the HATS and re-recorded by convolving the impulse response of the headphone to create the stimuli for the online test. The result from the laboratory listening test and online listening test were not directly comparable, but the result from the online listening test yielded consistent results across the partici- pants. This paper subsequently studied the headphone listening test procedure using the floor impact sound signal recorded by monaural microphone and HATS in the 6-storey model building in Building Research Institute, JAPAN. 2. RECORDING PROCEDURE AND THE TEST BUILDING

The sounds used for the subjective study were recorded in the 6-storey 2×4 model building and the CLT model building in Building Research Institute (BRI) shown in Figure 1. The impact sources used in the recording were the rubber ball which satisfies the requirement of standard heavy-impact source stated in JIS A 1418-2, ISO 10140-3: and 16283-2 [2-4], and the bang machine which satisfied the requirement of standard heavy-impact source in JIS A 1418-2. The excitation positions were the centre and the quarter length of diagonal of the slab and the drop height of the rubber ball was set as 10cm and 100cm. The bang machine was used as procedure states in the JIS standard. The micro- phone positions were directly below the excitation positions in the receiving room at a height of 0.9m when it was at the corner and 1.2m when it was in the centre from the floor level. The floor impact

i, orn inter.noise 21-24 AUGUST SCOTTISH EVENT CAMPUS ? O? ? GLASGOW

sound recordings on third floor (2×4), fourth floor (I-joist) and CLT model building were used in the listening test. Those recordings are later processed with MATLAB to add fade in and fade out.

i, orn inter.noise 21-24 AUGUST SCOTTISH EVENT CAMPUS ? O? ? GLASGOW

Figure 1. Picture of 6-storey 2x4 model building (left) and CLT model building (right) in Tsukuba, JAPAN

Figure 2. Floor impact sources: Bang machine and rubber ball.

3. THE LISTENING TEST

A schematic diagram of the listening test is shown in Figure 3. The interface shown on the iPad in Figure 4 was developed based on HTML5 and JavaScript for browser-based experience and adding background white noise. Eight participants were asked to listen to 36 monaural and stereo recordings (a total of 72 recordings) which were randomly played twice at three second intervals. The partici- pants were asked to answer 5-scale ratings on the degree of bother for each recording.

i, orn inter.noise 21-24 AUGUST SCOTTISH EVENT CAMPUS ? O? ? GLASGOW

Figure 3. Schematic diagram of the listening tests

Show the explanation screen Play the recording three times with 3 ‘Asked to answer 5- scale question second intervals skate ayer ache as foes ee hare ty arta a Complete wy > Repeat for 144 times Tatis Thankyou trae

Figure 4 listening test environment in anechoic chamber

4. RESULTS

4.1 Correlation coefficients of subjective ratings between stereo and monaural recordings for each participant The average of two attempts of 5- scale ratings on the degree of bother answers for each participant was calculated and used to calculate the correlation coefficients. The correlation coefficient on the tables were calculated using the Python library Pandas(ver.1.3.4) and it is Pearson correlation coeffi- cient, and rounded to the second decimal place. Table 1 shows the correlation coefficients between the ratings for monaural and stereo recordings calculated for each participant (sample size: 36 stim- uli). The correlation coefficient shows a minimum of 0.86 and a maximum of 0.93 which indicates monaural and stereo recordings are highly likely to be correlated. This suggested that the effect of spatial characteristics of the stimuli was not enough to change the relative relationship of the ratings within each participant.

Table 1 correlation coefficients between monaural and stereo answers for each participant

Participant # 1 2 3 4 5 6 7 8

Correlation Coefficient 0.93 0.92 0.93 0.88 0.87 0.86 0.93 0.92 4.2 Correlation coefficients of subjective ratings between stereo and monaural recordings for each stimulus Table 2 shows the correlation coefficients between the ratings for monaural and stereo recordings calculated for each stimulus (sample size: 8 participants). There are 5 recordings in which the corre- lation coefficient is equal to or less than 0.5. For those small correlation coefficient values, no ten- dency can be observed. It is possibly suggested that there were between-participant (or individual) differences in the effect of the spatial characteristics of the stimuli and that these differences increased or decreased depending on the combination of impact source, floor, excitation position, and micro- phone position.

Table 2. Correlation coefficients of each recording set Source Excitation Microphone 2×4 I-Joist CLT

Bang machine Centre Centre 0.92 0.85 0.90

Bang machine Corner Centre 0.77 0.63 0.41

Ball (1m height) Centre Centre 0.71 0.50 0.74

Ball (1m height) Corner Centre 0.94 0.64 0.52

Ball (0.1m height) Centre Centre 0.36 0.66 1.00

Ball (0.1m height) Corner Centre 0.68 0.71 0.70

Bang machine Centre Corner 0.80 0.91 0.81

Bang machine Corner Corner 0.83 0.90 0.75

Ball (1m height) Centre Corner 0.85 0.69 0.68

Ball (1m height) Corner Corner 0.91 0.75 0.33

Ball (0.1m height) Centre Corner 0.91 0.39 0.93

Ball (0.1m height) Corner Corner 0.75 0.65 0.74 4.3 Comparison in rating between monaural and binaural recordings The coefficient of determination of averaged over all participants for bang machine, ball(1m) and ball (0.1m) are shown in Figure 4. For the bang machine and rubber ball (1m), the rating by binaural recordings is higher than monaural ratings. For the rubber ball (0.1m) the rating by monaural is higher than binaural or equivalent. The ratings for the binaural recordings tend to shift toward a greater degree of bother than the monaural recordings when the ratings exceeded 3 (somewhat concerned). This means that a simplified evaluation using the monaural recordings will underestimate the degree of bother when the problem of bother is apparent to some degree.

i, orn inter.noise 21-24 AUGUST SCOTTISH EVENT CAMPUS ? O? ? GLASGOW

5

R² = 0.8524

4

Tire

R² = 0.918

3

ball 1m

Binaural

R² = 0.89

2

ball 10cm;

1

0

0 1 2 3 4 5

Monaural

Figure 5 Correlation coefficients and rating (average) 4. CONCLUSIONS

The paper studied the subjective evaluation test using the headphone in the anechoic chamber. A total of 72 recordings were tested for 8 participants. The correlation coefficient in the monaural and stereo answers from each participant shows a range between 0.86 and 0.93. This suggested that the partici- pants consistently answered the 5-scale ratings. The correlation coefficient for each recording sug- gested there is a slightly large range from 0.33 to 1. This indicates the participants may feel different to the sound. The ratings by monaural and binaural recordings are different as the rating from the binaural recording is slightly higher than the monaural recording for the bang machine and ball (1m) excitation sources. Further investigations are necessary for the relation between the rating framework and the listening test outcome, but the outcome from this paper suggests that there is a high possibility that stereo recordings and a headphone listening environment can be used for the listening test. 5. ACKNOWLEDGEMENTS

Authors gratefully acknowledge the funding from National Research Council Canada – New Begin- nings Initiative, and the Grand-in-Aid for Scientific Research by JSPS (KAKENHI) (20K14891). 6. REFERENCES

1. Markus Müller-Trapet , Iara Batista da Cunha, Jeffrey Mahn, Subjective studies on impact sound

in times of a pandemic – a comparison between laboratory study and an online listening test, Inter-noise (2021). 2. Japanese Standards Association, JIS A 1418-2:2019 Acoustics - Measurement of floor impact

sound insulation of buildings - Part 2: Method using standard, (2019). 3. ISO 10140-3:2010+A1:2015. Acoustics – Laboratory measurement of sound insulation of

building elements – Part 3: Measurement of impact sound insulation, (2014). 4. ISO 16283-2:2015 Acoustics – Field measurement of sound insulation in buildings and of

building elements – Part 2: Impact sound insulation, (2014).

i, orn inter.noise 21-24 AUGUST SCOTTISH EVENT CAMPUS ? O? ? GLASGOW