Precision of room acoustic modelling in open-plan offices Jukka Keränen, Pekka Saarinen, Valtteri Hongisto Turku University of Applied Acoustics, Acoustics Laboratory Joukahaisenkatu 7, FI-20520 Turku, Finland

ABSTRACT Room acoustic models can predict parameters that describe acoustic conditions in open-plan offices. An open-plan office of 81 m 2 was experimentally studied. Twenty-two room acoustic conditions were built using different combinations of absorption materials, and office screens. The conditions were measured according to ISO 3382-3. The conditions were modelled using Model A, a software based on raytracing, and Model B, a simple empirical model. The precision was assessed by the difference of predicted and measured single-number values (SNVs) of ISO 3382-3. The mean prediction accu- racy of L p ,A,S,4m was 3.2 dB and 2.0 dB for Models A and B, respectively. The mean prediction accu- racy of D 2,S was 1.2 dB and 2.4 dB for Models A and B, respectively. For Model A, the predicted and measured SNVs were similar in conditions with low sound absorption and low screen height. The agreement was worse in conditions with high sound absorption on the ceiling and walls, and over 1.6-m-high sound-absorbing screens. For Model B, the agreement of D 2,S was a little less precise in most conditions. L p ,A,S,4m with Model B was more precise than with Model A in conditions with high screens and high sound absorption, but less precise in other conditions.

1. INTRODUCTION

Speech is dominant sound source in open-plan and activity-based offices [1]. Distraction of speech can be controlled, e.g., using sound absorption materials, screens, and active masking sound systems. The effect of room acoustic improvements on the acoustic performance of the office can be measured according to ISO 3382-3 [2]. It defines five single-number-quantities: spatial decay rate of A- weighted sound pressure level (SPL) of speech, D 2,S [dB], A-weighted SPL of speech at 4-meter- distance from the speaker, L p,A,S,4m [dB], A-weighted SPL of background noise, L p,A,B [dB], distrac- tion distance r D [m], and comfort distance, r C [m]. The single-number values (SNVs) have also been predicted in case studies. The prediction methods were based on raytracing method [3-6] or empirical models [7,8]. Our study investigates the predic- tion accuracy of SNVs using the raytracing method [9] and the empirical model [7] in 22 room acous- tic conditions built in an open-plan office. The predicted values are compared to measured values, which have been already published [10]. 2. MATERIALS AND METHODS

2.1. Open-plan office and 22 studied conditions The studied open-plan office (81 m 2 ) involved 12 workstations (Figure 1). Twenty-two room acoustic conditions were built in the office. The conditions (Table 1) varied four parameters: ceiling absorption (high or low), wall absorption (high or low), screen absorption (`-´, high or low), and screen height (`-´, 1.3, 1.7, or 2.1 m).

The high ceiling absorption was implemented using 20 mm mineral wool tiles in the suspended ceiling (90% coverage) and low ceiling absorption using unperforated gypsum tiles. The high wall absorption was implemented by mounting 40 mm mineral wool panels on the wall surface (20% coverage). The low wall absorption was when the panels were not mounted. Two types of screens were used around the workstations: sound-absorbing office screens (high) and textile coated office screens (low). The screens were built to three different heights. The screen absorption and height are `-´ when the screens were removed.

Table 1: The investigated conditions. Description of sound absorption, screen height, and absorption coefficients for Mo del B.

Screen Condition height h [m] α c α f 1 high high high 2.1 0.85 0.8 2 high high high 1.7 0.85 0.7 3 high low high 2.1 0.85 0.6 4 high low high 1.7 0.85 0.5 5 high high high 1.3 0.85 0.6 6 high high low 1.3 0.85 0.4 7 high low high 1.3 0.85 0.4 8 high low low 1.3 0.85 0.2 9 low high high 2.1 0 0.8 10 low low high 2.1 0 0.6 11 low high high 1.7 0 0.7 12 high high - - 0.85 0.2 13 low low high 1.7 0 0.5 14 high low - - 0.85 0 15 low high high 1.3 0 0.6 16 low high low 1.7 0 0.5 17 low high low 1.3 0 0.4 18 low low high 1.3 0 0.4 19 low low low 1.7 0 0.3 20 low low low 1.3 0 0.2 21 low high - - 0 0.2 22 low low - - 0 0

Sound absorption Abs. coefficients ceiling / walls / screens ceiling / furniture

2.2. Measurements The room acoustic measurements are described in detail in [10]. Spatial decay of speech was deter- mined according to ISO 3382-3 [11] using a calibrated omnidirectional loudspeaker that produced loud pink noise. The SPL of the pink noise was measured in six workstations that were on two meas- urement paths (Figure 1). The SNVs were determined individually for both paths in all the 22 condi- tions. The mean SNVs were determined for all the 22 conditions.

Figure 1: The open-plan office with 12 workstations. The sound source was in the corner (O). The

measurement positions (  ) were in six workstations. Two measurement paths were used.

2.3. Model A A simulation software [9] based on raytracing method was used to predict the spatial decay of speech in the workstations (Figure 1). The SNVs were determined on both paths in the same way as in the measurements (Sec. 2.2).

The first model was created for condition 22 (low ceiling absorption, low wall absorption, no screens). The dimensions and locations of furniture and other midsize objects in the room were precisely measured and replicated in the model. The sound absorption coefficients were defined for the boundary surfaces and furniture that were the same in all the predicted conditions. This model was used to “calibrate” the predictions by comparison to measured results of SPL of speech and reverberation time in condition 22. The other conditions were modeled by creating variations of the first model. The variations involved the additional absorption materials, and the screens with alterna- tive absorption and height. The absorption coefficients were obtained from laboratory measurements and literature [10]. The sound source and the receivers were exactly in the same positions in the model as in the measurements.

2.4. Model B A simple empirical model has been developed for SNVs in open-plan offices [7]. The Equations (1) and (2) were determined for L p,A,S,4m and D 2,S . L p,A,S,4m was determined by

𝐿 𝑝,A,S,4m = 𝐿 𝑝,𝐴,𝑆,1𝑚 −3ℎ−0.1𝑊−4.6𝛼 𝑐 −0.8𝛼 𝑓 , (1)

where L p,A,S,1m =57.4 dB, h [m] is the screen height, W [m] is the room width, the α c is the ceiling absorption coefficient and α f is the apparent absorption coefficient of the furniture. D 2,S was deter- mined by

ℎ

𝐿

𝐷 2,S = 8

𝐻 + 4𝛼 𝑐 + 1.7𝛼 𝑓 , (2)

𝐻 + 0.16

where H [m] is the room height, and L [m] is the room length. The absorption coefficients α c and α f are shown in Table 1.

2.5. Prediction accuracy The mean of absolute prediction accuracy of both SNV was determined by

1

22 σ ห𝑥 𝑝 −𝑥 𝑚 ห 22 𝑛=1 , (3)

𝑥ҧ =

where x is either D 2,S or L p,A,S,4m . Standard deviation was determined using its general definition.

3. RESULTS

Figure 2 shows the measured L p,A,S,4m and D 2,S together with values predicted using models A and B in the 22 conditions.

The mean of absolute prediction accuracy for L p,A,S,4m and D 2,S is presented in Table 2. The abso- lute prediction accuracy of L p,A,S,4m varied within 0.1–9.1 dB with Model A, and within 0.0–5.5 dB with Model B. The absolute prediction accuracy of D 2,S varied within 0.1–3.7 dB with Model A, and within 0.1–4.0 dB with Model B.

Table 2: The mean of absolute prediction accuracy for L p,A,S,4m and D 2,S using Models A and B. The standard deviation is presented i n parentheses.

SNV Model A Model B

L p,A,S,4m 3.2 (2.5) 2.0 (1.2)

D 2,S 1.2 (1.0) 2.4 (1.1)

Figure 2: The measured and predicted L p,A,S,4m and D 2,S in the 22 conditions for Models A and B.

Lyasam [4B] =Measured ‘© Model A ‘2 Model B

The results are the mean values of both paths. 4. DISCUSSION

Models A and B require accurate information on the open-plan office. Surface materials, especially sound-absorbing, room dimensions, and sizes of screens and other large furniture need to be measured or determined from design plans. Neither model can guarantee precise results since inaccuracy of SNVs may be very large in individual office. In our study, the largest difference between predicted and measured SNV was -9.1 dB L p,A,S,4m (Model A, condition 1) and 4.0 dB D 2,S (Model B, condition 1). The mean of absolute prediction accuracy for L p,A,S,4m was 3.2 dB with Model A, and 2.0 dB with Model B. The mean of absolute prediction accuracy for D 2,S was 1.2 dB with Model A, and 2.4 dB with Model B. The L p,A,S,4m and D 2,S predicted using Model A agreed well with measured SNVs in conditions with low sound absorption and low screen height. The agreement was worse in conditions with high sound absorption on the ceiling and walls, and over 1.6-m-high sound-absorbing screens. The agreement of L p,A,S,4m predicted using Model B was similar in most of the conditions. The agree- ment of D 2,S predicted using Model B was less precise than Model A in conditions with low sound

Dz» (4B) 12345678 9 101ni21318151617 18192021 Condition =Measured ‘Model A ©Model B

absorption and lower screens. However, the measurements also have uncertainty. A recent Round- Robin test in an open-plan office presented reproducibility standard deviations 0.3 dB and 1.1 dB for D 2,S and L p,A,S,4m , respectively [12]. The room acoustic predictions enable assessing the effect of sound-absorbing materials and high screens on the SNVs. Model B is more reasonable for quick predictions, e.g., in design meetings because the method is remarkably faster than Model A. Model A is more detailed, and rather precise when used meticulously. Therefore, Model A is suitable for research purposes where more time can be allocated for creating the model geometries and determining all the details. 5. ACKNOWLEDGEMENTS

This study was a part of project “Active Work Space” that was funded by Academy of Finland [Grant 314788]. 6. REFERENCES

1. Yadav, M., Cabrera, D., Kim, J., Fels, J., de Dear, R. Sound in occupied open-plan offices: Ob-

jective metrics with a review of historical perspectives. Applied Acoustics, 177 , 107943 (2021). 2. International Organization for Standardization, Acoustics – measurement of room acoustic pa-

rameters – Part 3: Open plan offices . ISO 3382-3 (2022). 3. Rindel, J.H., Christensen, C.L. Acoustical simulation of open-plan offices according to ISO 3382-

3. Proceedings of Euronoise 2012 , Prague, Czech Republic, 10-12 June 2012. 4. Rindel, J.H. Open plan office acoustics – multidimensional optimization problem. Proceedings

of DAGA 2018 , Münich, Germany, 19-22 March 2018. 5. Jagla, J., Noé, N., Schmich-Yamane, I., Chevret, P. A hybrid method for open plan offices acous-

tics prediction using beam and particle tracing. Proceedings of Euronoise 2015 , Maastrich, The Netherlands, 31 May – 3 June 2015. 6. Pop, C.B., Rindel, J.H. Perceived speech privacy in computer simulated open-plan offices. Pro-

ceedings of Internoise 2015 , Rio de Janeiro, Brazil, 7-11 August 2015. 7. Keränen, J., Hongisto, V. Prediction of the spatial decay of speech in open-plan offices. Applied

Acoustics , 74 , 1315-1325 (2013). 8. Lüthi, G., Desarnaulds, V. Analysis of open plan acoustic parameters based on swiss and inter-

national databases of in situ measurements. Proceedings of ICSV27 , Prague, Czech Republic, 11- 16 July 2021. (Virtual) 9. Odeon A/S, Odeon Room Acoustics Software, User’s Manual, Version 14 (2018). 10. Keränen, J., Hakala, J., Hongisto, V. Effect of sound absorption and screen height on spatial decay

of speech – experimental study in an open-plan office. Applied Acoustics, 166 , 107340 (2020). 11. International Organization for Standardization, Acoustics – measurement of room acoustic pa-

rameters – Part 3: Open plan offices . ISO 3382-3 (2012). 12. Hongisto, V., Keränen, J., Labia, L., Alakoivu, R. Precision of ISO 3382-2 and ISO 3382-3 – A

Round-Robin test in an open-plan office. Applied Acoustics , 175 , 107846 (2021).