Comparing the Respirable Aerosol Concentrations and Particle Size Distributions Generated by Singing, Speaking and Breathing

26 The severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) pandemic has resulted in an 27 unprecedented shutdown in social and economic activity with the cultural sector particularly severely 28 affected. Restrictions on performance have arisen from a perception that there is a significantly higher 29 risk of aerosol production from singing than speaking based upon high-profile examples of clusters of 30

COVID-19 following choral rehearsals. However, no direct comparison of aerosol generation from 31 singing and speaking has been reported. Here, we measure aerosols from singing, speaking and breathing 32 in a zero-background environment, allowing unequivocal attribution of aerosol production to specific 33 vocalisations. Speaking and singing show steep increases in mass concentration with increase in volume 34 (spanning a factor of 20-30 across the dynamic range measured, p<1×10 -5 ). At the quietest volume (50 35 to 60 dB), neither singing (p=0.19) or speaking (p=0.20) were significantly different to breathing. At the 36 loudest volume (90 to 100 dB), a statistically significant difference (p<1×10 -5 ) is observed between 37 singing and speaking, but with singing only generating a factor of between 1. Health Organisation (WHO), with in excess of 21.5 million cases and 767,000 deaths reported worldwide 50 by 17 st August 2020. 4 Early in the pandemic, clusters of COVID-19 were considered to have arisen in 51 several choirs around the world. 5,6 This rapidly led to many governments restricting or suspending 52 singing. Concerns that woodwind and brass instruments might also be responsible for virus spread led to 53 similar restrictions on the playing of wind instruments. Consequently, large sections of the cultural sector, 54 along with religious institutions and educational establishments, were unable to rehearse and perform, 55 resulting in profound artistic, cultural, spiritual, emotional and social impacts. The livelihoods of many 56 performers have been jeopardised, and the viability of established institutions remains threatened. The 57 economic impact to the United Kingdom (UK) from this sector alone has been substantial, costing the 58 UK economy hundreds of millions in lost tax revenue, usually derived from the £32.2 billion cultural 59 purse. 7 60 Respiratory particulate matter is expelled during human exhalatory events, including breathing, speaking, 62 coughing and sneezing. [8][9][10] The flux generated is proportional to the amplitude of phonation in speech. 11 63 These actions release a plume of material containing particles of varying size, ranging from macroscopic 64 mucosalivary droplets originating from the oral cavity and pharynx, to microscopic aerosols released by 65 the small airways of the lungs. 8,9,11,12 Traditionally, the division between droplets, which are considered 66 to be of sufficient mass to sediment due to gravity, and aerosols, which remain airborne, is defined 67 arbitrarily at 5 µm diameter. 13,14 However, particle composition and environmental properties like 68 temperature, humidity and airflow influence the biophysical mechanics of the material released and the 69 extent of transport. 13 participants performing a range of activities including singing but have struggled to accurately quantify 86 aerosol and droplets because of the large number of background particulates in the environment. This 87 study is the first peer-reviewed study that explores the relative amounts of aerosols and droplets (up to 88 20 µm diameter) generated by a large cohort of 25 professional performers completing a range of 89 exercises including breathing, speaking, coughing and singing in the clean air environment of an 90 operating theatre with laminar flow ventilation. Measurements of particle number concentration alone 91 would be insufficient to determine the total amount of viral material capable of being transmitted: the 92 total mass of particulate matter produced may be a key factor in assessing the potential risk. Thus, 93 measurements of particle size distributions, as well as concentration, are used to assess the mass 94 concentration. 95

96
Overview of the Cohort of Professional Singers and the Study Design. The cohort of 25 professional 97 singers perform a broad range of genres, including musical theatre (6), choral (5), opera (5), and other 98 genres: gospel (2), rock (2), jazz (2), pop (1), actor with singing interest (1) and soul (1). 6 identified 99 their voice-type as soprano or mezzo-soprano, 7 as alto, 5 as tenor and 7 as bass or baritone. Aerosols 100 and droplet concentrations were measured with an Aerodynamic Particle Sizer (APS, 500 nm -20 µm) 101 in an operating theatre with each participant and researcher required to wear appropriate personal 102 protective equipment. The high air exchange rate, filtration and laminar air flow reduced the pre-existing 103 particle background number concentration to zero cm -3 , enabling the unique and extremely sensitive 104 measurements described. Thus, any particles detected were directly attributable to participant activity, 105 with particle concentrations returning to zero cm -3 between periods of singing, speaking and breathing. 106 Temperature and relative humidity were typically 20 o C and 45%, respectively. 107 108 A standard operating procedure was adopted (see Methods), covering 12 activities over ~1 hour, with 109 each activity involving up to 5 repeat actions, with a 30 s pause between each. These activities included 110 breathing, coughing, singing single notes ("/ɑ/") at different pitches, and speaking and singing the 111 "Happy Birthday" song at different volumes. At the beginning of each action, participants stepped 112 forward to the funnel (Fig. 1a) such that the dorsum of the nose was aligned to the plane of the base of 113 the cone. Participant position relative to the funnel was monitored to ensure consistency (within 10 cm 114 of the sampling tubes) across all measurements (Extended Data Fig. 1). As in previous studies, 9,10 we 115 report concentrations sampled through the collection funnel, which allows comparison of particle 116 emission rates on a relative basis between activities. In reality, particle concentrations will become 117 rapidly diluted once particles are exhaled, leading to strong spatial variations. 118 119 Aerosol Number Concentrations from Singing Compared with Other Expiratory Activities. A 120 sequence of measurements made with one APS for one performer is reported in Fig. 1 activities (5 repetitions of each). The zero-background is clearly apparent between measurements. 127 128 A complete analysis of the time-averaged total particle number and mass concentrations for all 25 129 participants is reported in Fig. 2. The statistical analysis is described in Methods and the absolute results 130 summarised in Extended Data Tables 1 and 2; data normalised to the aerosol concentration from speaking 131 at 70-80 dB are compared in Extended Data Fig. 2 and Table 3. The distribution of aerosol number 132 concentration generated across all participants is assumed to be log-normal, consistent with the data 133 presented in a previous publication; concentrations must always be positive-valued and a small number 134 of individuals generate a significantly larger aerosol flux than the median. 10 This is particularly apparent 135 for breathing, where measurements from individuals span almost three orders of magnitude. Indeed, 4 136 participants produced more aerosol in number concentration while breathing than while speaking at 90-137 100 dB. The reproducibility of concentration from singing a single note (70-80 dB) is not only apparent 138 in single participant data (Fig. 1b), but also across the cohort with median concentrations in good 139 agreement (0.83 and 0.91 cm -3 at beginning and end, respectively). At the lowest volume (50-60 dB), 140 neither singing (p=0.19) or speaking (p=0.20) were significantly different in particle production to 141 breathing, with median number concentrations of 0.10, 0.19 and 0.28 cm -3 for speaking, singing and 142 breathing, respectively. In the mixed model, compared to speaking, singing generates a statistically 143 significant (p < 1×10 -5 ) enhanced aerosol number concentration, although this enhancement is small 144 relative to the much larger changes associated with increase in volume (p < 1×10 -5 ). Aerosol number 145 concentration increases by a factor of 10-13 as volume increases from 50-60 dB to 90-100 dB, suggesting 146 that shouting should be associated with little difference in risk to singing at loud volume. singing, speaking and breathing generate aerosol particles of different size cannot be inferred by 171 comparing particle number concentrations alone. Instead, we must compare the aerosol size distributions 172 from these activities. Previously, two overlapping modes in the size distribution of particles from 173 speaking and coughing have been identified. 9,10 These have been attributed to distinct processes in this 174 expiration process. The mode of lowest size is generated in the lower respiratory tract with a second 175 mode generated in the region of the larynx, expected to be the most important in voicing. Figure 4 reports 176 the variation in mean number concentrations with particle size averaged over the 25 participants and 177 includes the fitted distribution from Johnson et al. 9 reported from a cohort of 15. Our distribution for 178 speaking and singing is in excellent agreement with the shape of the distribution reported by Johnson et 179 al. for particles larger than 800 nm diameter. Although the absolute concentrations are a factor of ~6 180 larger in our measurements, it should be recognised that the absolute value carries little meaning, 181 reflecting only the instantaneous value recorded by the APS from the sampling funnel, which will depend 182 on the sampling specifications. 10 183 184 Measured size distributions for speaking and singing were fitted to bimodal lognormal distributions. The 185 fits all gave the similar mean diameters and variance for both modes, further supporting the conclusion 186 that speaking and singing can be treated similarly (Extended Data Table 5). However, both vocalisations 187 generate larger particles than breathing: although the size distribution from breathing is well-represented 188 by a bimodal lognormal distribution, the larger mode is shifted to a smaller diameter and has a narrower 189 variance than for speaking and singing. Discussion. This study demonstrates that the assessment of risk associated with the spread of SARS-210 CoV-2 in large groups due to respirable particles from speaking and singing should consider the number 211 and mass concentrations of particles generated by these activities. The statistically significant, yet 212 relatively modest differences detected between the type of vocalisation at the loudest volume studied, 213 are eclipsed by the effects of volume on aerosol production, which varies by more than an order of 214 magnitude from the quietest to loudest volume studied, whether speaking or singing. By contrast, the 215 number of particles produced by breathing covers a wide range (spanning from quiet to loud volume 216 speaking and singing) but has a size distribution shifted to smaller particle sizes, in principle mitigating 217 some of the potential risk associated with the wider emission range. 218

219
We also find that a minority of participants emitted substantially more aerosols than others, sometimes 220 more than an order of magnitude above the median, consistent with the long-tail of a log-normal 221 distribution when viewed in linear-concentration space. This observation is consistent with a previous 222 study. 10 However, the highest emitters were not consistently the highest across all activities, suggesting 223 the magnitude of emission from an individual may be highly activity specific. It is unclear why some 224 participants emit substantially more than others, and further studies are required to better characterise the 225 variability of aerosol emission across the population, as well as the consistency of emission from an 226 individual over time. For certain vocal activities and venues, amplification may be a practical solution to reduce the volume 240 of singing by the performers. Based on the differences observed between vocalisation and breathing and 241 given that it is likely that there will be many more audience members than performers, singers may not 242 be responsible for the greatest production of aerosol during a performance, and for indoor events 243 measures to ensure adequate ventilation may be more important than restricting a specific activity.

Human subjects 319
The Public Health England Research Ethics and Governance of Public Health Practice Group (PHE 320 REGG) approved this study and all research was performed in accordance with relevant guidelines and 321 regulations of the Ethical Review Board. We recruited 25 healthy volunteers (12 males and 13 females, 322 ranging in age from 22 to 57 years old (mean 38, SD +/-9.8) through contact and collaboration with the 323 entertainment industry. Informed consent was obtained from all participants prior to study participation. 324 All participants completed a pre-screening questionnaire including age, gender, professional status, 325 singing training history and COVID 19 symptom status to fulfill inclusion/exclusion criteria. Only 326 participants who self-reported no symptoms of COVID-19 and who had normal temperatures on the day 327 of attendance were included. Each participant's weight, height and peak flow rate were measured before 328 the aerosol measurements. Body mass index was calculated from the height and weight measurement. 329 330

Aerosol Measurements 331
Measurements were performed simultaneously with two APS instruments (TSI 3321) and one Optical 332 Particle Sizer (OPS, 0.3 -10 μm, TSI 3330) sampling from the same custom-printed funnel. A 333 comparison of measurements between the two APS instruments was linear, with a slope that deviated 334 from 1 owing to different sensitivities of the instruments (Extended Data Fig. 5). The OPS detected 335 significantly more particles than the APS (up to a factor of 2), a consequence of the lower size detection 336 limit of the OPS (to 300 nm) compared to the APS (to 500 nm) (Extended Data Fig. 6). Including these 337 smaller particles in our analysis significantly increases the number concentration but does not 338 significantly change the particulate mass concentrations from the expiratory activities. 339

340
The sampling funnel was 3D printed from PLA (1.75 mm filament) by a RAISE3D Pro2 Printer 341 (3DGBIRE). The funnel was 150 mm wide, 90 mm deep with 3 ports at the neck for sampling aerosol 342 into up to three aerosol instruments (some combination of APSs and OPSs). All tubing was conductive 343 silicone and 130 cm in length (TSI Inc., product number 3001788, inner diameter 0.19 inch, outer 344 diameter 0.375 inch). 345 Participants voiced /ɑ/ (the vowel sound in 'saw') for 10 s at 70-80 dB in close proximity to the funnel 349 followed by 30 s of nose breathing and standing 2 m away from the funnel, repeated four more times in 350 succession. The participant repeated the series of five /ɑ/ vocalisations at the same amplitude using 351 feedback from a decibel meter. Soprano/mezzo soprano singers sang note F4, alto note D4, tenor note 352 F3 and baritone/bass note C3. After each set of experiments participants were asked to take a sip of water. Participants sang the "Happy Birthday" song to "Dear Susan" for 20 s at 50-60 dB followed by 30 s of 366 nose breathing and standing 2 m away from the funnel, repeated four more times in succession. The 367 participants then repeated this sequence at 70-80 dB and at 90-100 dB. Soprano/mezzo soprano singers 368 sang in B flat major (starting note F4, top note F5), alto in G major (starting note D4, top note D5), tenor 369 in B flat major (starting note F3, top not F4) and baritone/bass in F major (starting note C3, top note C4). 370

Breathing experiments 372
Participants breathed for 10 s inhaling through the nose and exhaling through an open mouth in a non-373 forced "quiet" fashion, then stood 2 m away from the funnel for 30 s in between each breathing 374 experiment and repeated four more times. An additional set of five breathing measurements were 375 conducted in similar fashion but where the participants inhaled through the nose and exhaled out of the 376 nose in a "quiet" fashion. 377 participant repeated the series of five /ɑ/ vocalisations at the same amplitude using feedback from a 382 decibel meter. Soprano/mezzo soprano singers sang note F4, alto note D4, tenor note F3 and 383 baritone/bass note C3. 384

385
Coughing 386 Participants were asked to cough into the funnel once, stand 2 m away for 30 seconds and then repeat 387 this process two more times. participants. There is no correlation of concentration with BMI (R-Squared is 0.3449 for breathing and 477 0.0004 for singing 90-100 dB). b) Variation of aerosol number concentrations generated by breathing 478 (red squares) and singing (90-100 dB, black squares) with peak flow rate across 25 participants. There is 479 no correlation of concentration with peak flow (R-Squared is 0.0011 for breathing and 0.0075 for singing 480 90-100 dB). Note that the clustering of data in part b) represents gender differences: males have a higher 481 peak flow rate than females. 482 Extended Data Figure 5: Comparison of measurements from two APS instruments across 8 participants. 485 Both instruments are linearly correlated, although the slope is less than 1 because the second APS 486 instrument (APS2) is less sensitive than the first (APS1