Proceedings of Meetings on Acoustics

Proceedings of Meetings on Acoustics Volume 19, 2013 http://acousticalsociety.org/ ICA 2013 Montreal Montreal, Canada 2-7 June 2013 Noise Session 4aNSa: Effects of Noise on Human Performance and Comfort I 4aNSa6. Prediction of the effectiveness of a sound-masking system in an open-plan office including the Lombard Effect Yizhong Lei* and Murray Hodgson *Corresponding author's address: Mechanical Engineering, The University of British Columbia, Room 2160, CIRS, UBC, Vancouver, V6T1Z4, British Columbia, Canada, light.lei@hotmail.com Sound masking can improve speech privacy in rooms by increasing background-noise levels that mask distracting speech sounds. The Lombard Effect indicates that an increase in background-noise level can increase talker voice levels, reducing speech privacy and the benefit of a sound-masking system. To investigate this, a model of an existing open-plan office was created in CATT-Acoustics and validated. The model was used to predict speech-transmission index (STI) and the effectiveness of a sound-masking system, without and with the Lombard Effect, described by the existing Lombard Voice Model. Predictions were made for ambient-noise levels of 30, 40 and 45 dba, at various distances from a primary talker, and for 0-4 secondary talkers. With 30-dBA ambient noise, STIs at 1 and 4 m varied with the number of talkers from 0.67-0.91 and 0.23-0.62 without the sound-masking system, from 0.58-0.70 and 0.13-0.25 with it but ignoring the Lombard Effect, and from 0.64-0.73 and 0.18-0.27 with the Lombard Effect. The Lombard Effect reduced the benefit of the sound-masking system by 14-67% and 5-50%. With higher ambient noise, the system is less effective; the Lombard Effect can cancel its benefit, resulting in increased STI (decreased privacy) with the system operating. Published by the Acoustical Society of America through the American Institute of Physics 2013 Acoustical Society of America [DOI: 10.1121/1.4800263] Received 21 Jan 2013; published 2 Jun 2013 Proceedings of Meetings on Acoustics, Vol. 19, 040111 (2013) Page 1

INTRODUCTION Sound masking can improve speech privacy (SP) in rooms by increasing background-noise levels that mask distracting speech sounds. Sound masking was considered necessary in the office-design guidelines of Hardy [1]. However, the Lombard Effect elucidated by Lombard [2], indicates that an increase in background-noise level can increase talker voice levels, reducing SP and the benefit of a sound-masking system. This paper discusses the effect of the Lombard Effect on the benefit provided by sound masking in an open-plan office, by way of physical measurement and prediction. An existing open-plan office in the University of British Columbia was investigated, and a model of this open-plan office was created in CATT-Acoustics and validated. The model was used to predict speech-transmission index (STI) and the effectiveness of a sound-masking system, without and with the Lombard Effect, described by the existing Lombard Voice Model. Predictions were made for ambient-noise levels of 30, 40 and 45 dba, at various distances from a primary talker, and for 0-4 secondary talkers. FIGURE 1. The BSS Lab in the CIRS building in UBC. OPEN-PLAN OFFICE AND EXPERIMENTAL WORK The investigation was based on an existing open-plan office the BSS Lab (a student study area) in the naturally-ventilated CIRS building at UBC (see Figure 1). The structural shape of the BSS Lab is approximately a rectangular parallelepiped, with length = 15.3 m, width = 10.2 m and height = 3.7 m. The surfaces and furnishings are mostly sound-reflective. L-shaped desks are partially surrounded by 1.25-m-high partitions. Acoustical measurements were made in the unoccupied Lab. Background-noise measurement was performed using a Rion NA-28 sound-level meter. Impulse-response (IR) measurements were made using WinMLS installed on an Acer laptop computer. The sound-level meter was connected to the input of the laptop; an omnidirectional dodecahedral loudspeaker array which generated maximum-length sequences was connected through a Bryston 4B power amplifier to the output. From the filtered IRs, the 125- to 8000-Hz octave-band sound-decay curves and reverberation times (T30) were calculated. Measurements were also made of total A-weighted sound-pressure levels (SPLs) corresponding to a Casual voice using a speech-source with directional-radiation characteristics similar to those of a human talker. Figure 2 shows the Lab floor-plan and measurement configuration. FIGURE 2. Floor-plan of the BSS Lab, showing the speech-source ( ) positions, the receiver lines ( ) and positions ( ) for speech-source sound-pressure-level measurement, and the receiver positions ( ) for reverberation-time measurement. Proceedings of Meetings on Acoustics, Vol. 19, 040111 (2013) Page 2

TABLE 1. Ambient background-noise levels in the unoccupied BSS Lab with the SMS off and on. Bands (Hz) 125 250 500 1000 2000 4000 8000 A-weighted total SMS off (db) 38.6 34.5 25.0 21.6 17.7 14.3 13.0 30.2 SMS on (db) 49.4 47.6 44.2 41.1 36.6 31.8 22.5 46.4 FIGURE 3. (a) Measured and predicted reverberation times in the unoccupied BSS Lab; (b) Variation of Casual-voice speech SPL with distance along two lines in the unoccupied BSS Lab. Table 1 shows the unweighted octave-band, and the A-weighted, total ambient background-noise levels (BNL) measured in the BSS Lab. The total A-weighted level is a very low 30.2 dba. Figure 3(a) shows the T30 results. Reverberation time ranged from 0.44 s at 250 Hz to 0.60 s at 1000 Hz. The average RT at 500 and 1000 Hz was 0.54 s. Figure 3(b) shows the total, A-weighted speech levels along the two measurement lines, on average levels decreased by approximately 5 dba per doubling of distance. PREDICTION METHODS In order to analyze the effect of a SMS, and of the Lombard Effect, the CATT-Acoustics software was used to create a model of the BSS Lab and predict speech-transmission index (STI) in it. CATT-Acoustics is a roomacoustical prediction tool based on ray-tracing and the image- source method. Room structures and surfaces were characterized by values of their absorption and diffuse-reflection coefficients. BNL was defined in octave bands as the measured values in Table 1. The room contains a primary talker speaking to a listener at which the acoustical conditions are evaluated, as well as a number of secondary talkers. The talker voice level ( Casual [3]) was defined. The total noise level (Ln) results from BNL and the reverberant field of the voices of the secondary talkers. The speech level (Ls) is the result of the direct field of the voice of the primary talker. Echograms and derived acoustical parameters reverberation time (T30, for an omnidirectional source), speech-source SPLs (speech levels, Ls) and STI were predicted. The model of the BSS Lab was created by modeling the major surfaces of the room and its furnishings, and defining their acoustical characteristics. Figure 4 shows the geometric model. Table 2 shows the absorption and diffuse-reflection coefficients used in the model, taken from Refs. [4] and [5]. Diffuse-reflection coefficients were estimated based on material shapes, surface rigidity and roughness [6]. TABLE 2. Absorption/diffuse-reflection coefficients of materials used in prediction [4, 5, 6]. Bands (Hz) 125 250 500 1000 2000 4000 Wood 0.31/0.30 0.33/0.30 0.14/0.30 0.10/0.30 0.10/0.30 0.14/0.30 Desktop 0.06/0.35 0.07/0.35 0.09/0.35 0.09/0.35 0.08/0.35 0.07/0.35 Partition 0.45/0.85 0.44/0.85 0.62/0.93 0.92/0.87 0.97/0.88 0.99/0.94 Glass 0.35/0.10 0.25/0.10 0.18/0.10 0.12/0.10 0.07/0.10 0.04/0.10 Carpet 0.09/0.35 0.08/0.55 0.21/0.78 0.26/0.88 0.27/0.92 0.37/0.93 Plasterboard 0.03/0.10 0.15/0.10 0.10/0.10 0.05/0.10 0.04/0.10 0.05/0.10 Panel 0.12/0.60 0.09/0.70 0.09/0.82 0.09/0.72 0.08/0.73 0.09/0.73 Frame 0.06/0.10 0.07/0.15 0.09/0.20 0.09/0.25 0.08/0.28 0.07/0.30 Porous material 0.47/0.35 0.65/0.35 0.75/0.35 0.84/0.35 0.83/0.35 0.81/0.35 Seat 0.19/0.30 0.37/0.30 0.56/0.30 0.67/0.30 0.61/0.30 0.59/0.30 Proceedings of Meetings on Acoustics, Vol. 19, 040111 (2013) Page 3

FIGURE 4. The CATT-Acoustics model of the BSS Lab. Model verification is necessary to ensure the validity of the model and the accuracy of its predictions. Figure 3(a) shows the T30 comparison. The differences varied from -3.3% at 1000 Hz to +8.5% at 500 Hz. The model is acceptable for use to investigate the effects of a SMS and the Lombard Effect. SOUND-MASKING SYSTEM The benefit of an SMS was first investigated for a BNL value of 30 dba as in the existing BSS Lab, as well as for higher values of 40 and 45 dba, more typical of mechanically-ventilated buildings. Even higher values of BNL were not investigated since they are higher than the optimal masking-noise level of 45-48 dba [7]). System Design and Modelling For an open-plan office like the BSS Lab (with screens 1.25-m high, some reflective surfaces and moderate furniture absorption), the masking-noise levels in third-octave bands should be as shown in Table 3 [7]. Note that these levels correspond to an A-weighted total level of 46.5 dba. Figure 4 shows the assumed locations of the SMS loudspeakers. The distances between them are 3.0 m in width, and 5.6 m in length. The output levels (in terms of free-field levels at 1 m) of the loudspeakers are shown in Table 4. The uniformity of SPLs in the office area was predicted to verify uniformity. Eight receiver positions, 1.2-m high, were defined in the room model, as shown in Figure 5. Since the BSS Lab is approximately symmetrical with respect to the length and width axes, the receivers were only located in one-quarter of the area. Receivers 1 to 6 were placed in the seats of workstations, Receiver 7 was in the main pathway and Receiver 8 was at the room center. TABLE 3. Ideal SMS masking-noise SPL spectrum and the acceptable variability (tolerance) [7]. Third-octave band (Hz) SPL (db) Tolerance (db) Third-octave band (Hz) SPL (db) Tolerance (db) 50 45.0 (+4/-6) 800 37.3 (±2) 63 45.0 (+4/-5) 1000 36.0 (±2) 80 45.0 (+4/-4) 1250 34.7 (±2) 100 45.0 (±3) 1600 33.3 (±2) 125 45.0 (±3) 2000 32.0 (±2) 160 44.3 (±3) 2500 30.3 (±2) 200 43.7 (±2) 3150 28.7 (±2) 250 43.0 (±2) 4000 27.0 (±2) 315 42.0 (±2) 5000 23.7 (±2) 400 41.0 (±2) 6300 20.3 (±3) 500 40.0 (±2) 8000 17.0 (±4) 630 38.7 (±2) 10000 12.0 (+4/-6) TABLE 4. SMS loudspeaker octave-band output levels in terms of the free-field SPL at 1 m (Lpff1 in db). Bands (Hz) 125 250 500 1000 2000 4000 8000 Total Lpff1 (db) 54.0 53.5 48.5 45.5 41.8 36.7 26.7 59.5 Proceedings of Meetings on Acoustics, Vol. 19, 040111 (2013) Page 4

5 3 1 6 4 2 8 7 FIGURE 5. SMS prediction loudspeaker (dots) and receiver positions (numbered squares). 55 50 SPL (dba) 45 40 35 30 125 250 500 1000 2000 4000 Frequency (Hz) FIGURE 6. SMS SPLs predicted at eight receivers for uniformity verification and acceptable variation. Figure 6 shows the average and range of the SPLs due to the SMS at the eight receiver positions predicted in the BSS Lab. The results indicate that the SMS as designed increases the background noise to the desired levels, and that the uniformity of the levels is acceptable. After introducing the SMS into the BSS Lab, the average noise level increased from 30.2 to 46.4 dba, as shown in Table 1. Prediction STI predictions were made along receiver Line #1 (see Figure 2), across at source-receiver distances of 1, 2, 4, 6 and 8 m, and for 1, 2, 3, 4 and 5 (1 primary and 0, 1, 2, 3 and 4 secondary) talkers. In open-plan offices, Good or Excellent speech intelligibility (SI, STI 0.60) is required within a workstation. A typical distance between talker and listener in a workstation is 1 m. For people in different workstations, SP is more important than SI, so Good or Excellent SP (STI 0.45) is needed. The distances between seats in workstations in the BSS Lab are greater than 2 m. Figure 7 plots the STI results with one talker as a function of source-receiver distance across the BSS Lab. STI tends to decrease with distance but remained similar from 4 m. Thus, the STIs at distances of 1 and 4 m were chosen to evaluate the acceptability of SI and SP, respectively. Figures 8 (1 m) and 9 (4 m) plot the full prediction results as a function of the number of talkers and show the corresponding quality ratings. With the existing BNL=30.2 dba, SI within workstations is Excellent with 1 or 2 talkers and Good with more; SP between workstations is Poor with 1 talker and Good with more. With higher BNL=40 or 45 dba, SI within workstations remains Excellent or Good, except at the highest noise level with 5 talkers for which SI is Fair. Similarly, turning on the SMS increases the background-noise level to an acceptable level of 46.4 dba. Ignoring the Lombard Effect, this results in Good SI with 1-4 talkers and Fair SI with five. A background-noise level above 40 dba including that resulting from turning on the SMS impairs verbal communication within workstations; however, it improves SP between workstations, especially when the ambient background-noise level was low beforehand, and makes it Excellent. Proceedings of Meetings on Acoustics, Vol. 19, 040111 (2013) Page 5

Speech Transmission Index 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 SMS is off 0.1 SMS is on 0 0.5 1 2 4 6 8 Source-receiver distance (m) SI Excellent Good Fair Poor Bad SP Bad Poor Fair Good Excellent FIGURE 7. Predicted STI along receiver Line #1, across with one talker, without and with an SMS. 1.0 SI 0.9 STI 0.8 0.7 0.6 Excellent Good Fair BNL=30.2dBA BNL=40.0dBA BNL=45.0dBA SMS ON SMS ON with LE BNL=30.2dBA LE BNL=40.0dBA with LE BNL=45.0dBA with LE 0.5 1 2 3 4 5 Number of talkers FIGURE 8. Variation of STI along receiver Line #1, across with number of talkers in different conditions with distance=1m. STI 0.7 0.6 0.5 0.4 0.3 0.2 0.1 SP Poor Fair Good Excellent BNL=30.2dBA BNL=40.0dBA BNL=45.0dBA SMS ON SMS ON with LE BNL=30.2dBA with LE BNL=40.0dBA with LE BNL=45.0dBA with LE 0.0 1 2 3 4 5 Number of talkers FIGURE 9. Variation of STI along receiver Line #1, across with number of talkers in different conditions with distance=4m. Proceedings of Meetings on Acoustics, Vol. 19, 040111 (2013) Page 6

TABLE 5. Lombard-Effect parameters used in predictions. Lpff1,q (dba) asym (dba) xmid (dba) scale 52.4 29.5 70.0 7.5 LOMBARD EFFECT To predict talker voice levels considering the Lombard Effect, the Lombard Voice Model proposed by Hodgson, Steininger and Razavi [8] was used. This involved modeling the room and predicting talker voice levels using the Lombard Voice Model [8], and then modeling the room and the (primary and secondary) talkers, and predicting speech and noise levels, room echograms and corresponding STI values, using CATT-Acoustics. It models a room with an ambient non-speech background-noise level, BNL, and a primary talker speaking to a nearby primary listener at received speech level Ls. The room also contains a number of more distant secondary talkers. The model predicts the voice output level of all talkers using diffuse-field theory and a model of the Lombard Effect, given the room dimensions, absorption (or reverberation time), the number of secondary talkers, and the total backgroundnoise level Ln due to the combined reverberant field of the non-speech sources, possibly including an SMS, and the secondary talkers voices. The Lombard Voice Model equation is as follows [8]: asym L L db A 1 exp xmid Ln scale pff 1, n pff 1, q. (1) In Eq. (1), L pff1,n, asym, xmid and scale are Lombard Voice Model constants; the values, as shown in Table 5, were chosen from references, experience and historical data [8]. The reverberation times used in the Lombard Voice Model were the measured values instead of those predicted, as they were considered more accurate. Talker voice levels, determined using the Lombard Voice Model, were used to define the output of the speech source in the model of the BSS Lab and to predict STI at different receiver distances. The prediction results at 1 and 4 m are shown in Figures 8 (1 m) and 9 (4 m). With BNL=30.2dBA, consideration of the LE increases STI (decreases SP) by up to 0.03-0.06 (3-9%) to 0.64-0.73 at 1 m, and by 0.02-0.05 (3-22%) to 0.18-0.27 at 4 m, the effect increasing with the number of talkers. The benefit of the SMS (decreased STI) decreases by 14-67% from 0.09-0.21 to 0.03-0.18 at 1 m, and by 5-50% from 0.10-0.37 to 0.05-0.35 at 4 m. With higher ambient noise levels, such as BNL=40 or 45 dba, the increased noise level due to the SMS of course has less effect, as discussed above. However, with the Lombard Effect taken into consideration, the benefit decreases and can be canceled the Lombard Effect can almost cancel the benefit of the SMS, resulting in increased STI (decreased SP) relative to that with no SMS. With BNL=40 dba, for example, the benefit of the SMS (decreased STI) decreases by 15-67% from 0.06-0.13 to 0.02-0.11 at 1 m and by 6-38% from 0.08-0.16 to 0.05-0.15 at 4 m due to the LE. With BNL = 45 dba the corresponding decreases are 50-100% at 1 m and 0-50% at 4 m. In both cases the effects increase with the number of talkers. CONCLUSION The effects of ambient background noise, including its increase due to an SMS, were predicted for various numbers of talkers, without and with consideration of the Lombard Effect. Without considering the Lombard Effect, the SMS slightly impairs SI within workstations, but can significantly improve SP between workstations. The effects increase with the magnitude of the change in noise level due to the SMS, and with the number of talkers in the room. The Lombard Effect significantly reduces the benefit of an SMS on improving SP between workstation. The reduction in benefit decreases as the amount the SMS increases noise levels decreases; when the increase is small, the Lombard effect can cancel the estimated benefit. The results confirm that the Lombard Effect should be taken into consideration when designing an SMS and predicting its benefits. ACKNOWLEDGMENTS The authors thank Shira Daltrop, Nathan Willson, and Ahmed Summan for their valuable assistance in planning the tests and building the model described here. Proceedings of Meetings on Acoustics, Vol. 19, 040111 (2013) Page 7

REFERENCES 1. H. C. Hardy, A guide to office acoustics, Architectural Record, 121(2) 235-240 (1957). 2. E. Lombard, Le signe de l élévation de la voix, Annales des Maladies de l Oreille, du Larynx, du Nez et du Pharynx, 37, 101-119 (1911). 3. ANSI S3.5-1997, American National Standard Methods for Calculation of the Speech Intelligibility Index, Acoustical Society of America, New York (1997). 4. M. Long, Absorption Coefficients of Common Materials, Architectural Acoustics, London: Elsevier Academic Press (2006). 5. M. D. Egan, Sound Absorption Data for Common Building Material and Furnishings, in Architectural Acoustics, New York: McGraw-Hill, (1988). 6. CATT Acoustics, Version 8. Room Acoustics Prediction and Desktop Auralization: User s Manual (CATT-Acoustic, Sweden, 2002). CATT-Acoustic v8.0g, build 2, CATT, Gothenburg, Sweden (2007). 7. A. Taylor, Sound Masking Systems, Report to Atlas Sound, Hoover & Keith Inc. (2000). 8. M. Hodgson, G. Steininger and Z. Razavi, Measurement and prediction of speech and noise levels and the Lombard Effect in eating establishments, J. Acoust. Soc. Am., 121(4), 2023 2033 (2007). Proceedings of Meetings on Acoustics, Vol. 19, 040111 (2013) Page 8