Large Particle Emissions from Human Vocalization and Playing of Wind Instruments

Abstract

Humans emit large salivary particles when talking, singing, and playing musical instruments, which have implications for respiratory disease transmission. Yet little work has been done to characterize the emission rates and size distributions of such particles. This work characterized large particle (d_p > 35 μm in aerodynamic diameter) emissions from 70 volunteers of varying age and sex while vocalizing and playing wind instruments. Mitigation efficacies for face masks (while singing) and bell covers (while playing instruments) were also examined. Geometric mean particle count emission rates varied from 3.8 min^–1 (geometric standard deviation [GSD] = 3.1) for brass instruments playing to 95.1 min^–1 (GSD = 3.8) for talking. On average, talking produced the highest emission rates for large particles, in terms of both number and mass, followed by singing and then instrument playing. Neither age, sex, CO₂ emissions, nor loudness (average dBA) were significant predictors of large particle emissions, contrary to previous findings for smaller particle sizes (i.e., for d_p < 35 μm). Size distributions were similar between talking and singing (count median diameter = 53.0 μm, GSD = 1.69). Bell covers did not affect large particle emissions from most wind instruments, but face masks reduced large particle count emissions for singing by 92.5% (95% CI: 97.9%, 73.7%).

Short abstract

Aerosols and droplets are potential vectors for the transmission of airborne infectious disease; this work provides quantitative count and mass emission rates of large particles (d_p > 35 μm) for talking, singing, and playing wind instruments.

Introduction

The COVID-19 pandemic has motivated a growing body of research on human respiratory aerosol generation and disease transmission associated with activities like talking, singing, and playing musical instruments.¹⁻¹³ To date, most of this research has focused on smaller particles (aerodynamic particle diameter, d_p < 10 μm), which are relatively easy to count and size using conventional instruments like optical particle counters and aerodynamic particle sizers.¹⁴ Smaller particles are important because they can remain airborne for hours and travel distances beyond typical “close-contact” thresholds used in disease control guidance (∼2 m) and have been shown to transmit infectious pathogens in air.^7,15 However, larger particles (d_p > 10 μm) are also emitted during the same respiratory aerosol generation processes. Larger particles remain airborne for only seconds to minutes and are emitted at rates lower than those of their smaller counterparts; however, their mass emissions far outweigh those of smaller particles. Thus, larger particles have the potential to harbor more pathogens, on a per-particle basis, which could result in a more concentrated dose delivered to a given cell/receptor.¹⁶⁻¹⁸ Pathogens transported within large particles may also remain viable for longer periods of time than smaller ones due to their substantially longer evaporation times in typical room air conditions;⁶ these characteristics may also allow them to serve as vectors for fomite transmission after depositing onto a surface that is later contacted by a new host.^7,19 Additionally, some respiratory diseases are more commonly transmitted via larger particles than smaller ones.¹⁸ Therefore, it is important to characterize emission rates for all particle sizes.

Conventional aerosol counting and sizing instruments are limited in their ability to measure larger particles because such particles have a relatively high inertia in air (i.e., they experience high rates of aspiration and transmission losses during sampling).²⁰ Open-path laser instruments have been used to size larger airborne particles,^9,17 but such measurements are considered semiquantitative, especially under turbulent flow conditions (e.g., during vocalization). For example, Stadnytskyi et al.¹⁷ reported emissions rates of large and small particles emitted from 25 s of talking; however, these size-resolved results are considered semiquantitative because there was no effective way to equate scattered light intensity to particle size, since light intensity varied across the laser sheet. They referred to the top 25% of the brightest light scattering particles as large and the remainder as small. This measurement challenge has produced a gap in our understanding of large particle emissions from various human activities. Several previous studies have sought to characterize the full spectrum of aerosol sizes,^9,17,21−25 but these experiments were limited by small sample sizes (n < 12) and mainly focused on emissions from talking with little variation in the type of speech and articulation investigated. A recent study by Harrison et al.⁵ utilized a larger cohort of participants to investigate large particle emissions from speech (n = 43) and singing (n = 26) but did not consider emissions from any musical instruments or the effect of face coverings. The objective of this work was to characterize large respiratory aerosol (35 < d_p < 100 μm) and droplet (d_p > 100 μm) emissions from a panel of individuals (n = 70) of varying age and sex while talking, singing, and playing various wind instruments. A secondary objective was to examine the efficacy of face masks and instrument bell covers (i.e., single layered Spandex-like woven fabric covers that fit tightly around the instrument bell) at mitigating these emissions. The cutoff at 35 μm used here is based on the upper size limit of detection for the measurement techniques used in the Good et al.²⁶ and Volckens et al.²⁷ studies, which characterized small particle emissions (0.25 < d_p < 35 μm) from the same set of vocalization and instrument performances, and because particles smaller than 35 μm are difficult to detect using the low-magnification optical methods employed here.

Methods

Data Collection

Data were collected from a panel of healthy volunteers as part of the CSU Performing Arts Aerosol Study.^26,27 Participants were recruited in equal proportions of minors (aged 12–17 years at the time of participation) and adults (age 18 years and older at the time of participation), aiming for an equal distribution of participants across types of musical instruments and skill level. For the safety of staff and participants, people were excluded from the study if they were currently experiencing COVID-19 symptoms, had a COVID-19 diagnosis in the past month, or had close contact with an infected person in the previous 2 weeks (consistent with state and local quarantine protocols at the time). Participants brought in their own musical instruments and face masks, which varied in type from homemade cotton and other types of fabric materials to surgical-style masks. Upon arrival, participants provided written informed consent or assent if a minor (along with consent from their legal guardian), as detailed by the US regulatory guidelines for research on human subjects (CSU approval #20-10174H). They also provided information about their height, weight, age, and sex. Additional information on the study procedures can be found in previous publications.^26,27

Participants donned polypropylene disposable coveralls (McMasterCarr, IL) and hair nets to minimize particle emissions from their clothing and hair. Participants then entered a 3.45 m × 2.8 m × 2.45 m clean-room environment with low background particle (0.25 < d_p < 35 μm) concentrations (∼5 cm^–3) and a high, HEPA-filtered, air circulation rate (∼8.5 air changes per hour).^26,28 Temperature and humidity were recorded in the clean room using LabVIEW (National Instruments, TX, version 21.0, https://www.ni.com/en-us/shop/labview.html) instrument control and data acquisition software. Over the course of a 2-h measurement period, participants either sang or played their instrument (using provided sheet music) into an aerosol sampling inlet cone both with and without masks or bell covers, and everyone ended their sessions with a 4 min period of reading out loud (no mask) from a predetermined passage, as previously described.^26,27 Participants were asked to position their instrument bells or mouths (if talking or singing) right at the entrance plane of the inlet cone (Figures S1 and S2), so that all particle emissions were entrained into the cone’s sampled airflow (10 L·min^–1). However, as noted above, particles larger than ∼35 μm in aerodynamic diameter are likely to settle under the force of gravity prior to reaching the aerosol sizing equipment. Therefore, the lower half of the inlet cone was arrayed with seven 5 cm × 7.5 cm water-sensitive paper (WSP) cards (TeeJet Technologies Water Sensitive Paper Spraying Systems Co.). The card array was affixed to a single magnetic tray to streamline the process of setup and takedown, as shown in Figure S3. These cards undergo a permanent change in color when they come into contact with water (see Figure 1A), allowing each liquid particle that lands on the WSP surface to be imaged, sized, and counted under a microscope.

Figure 1.

Example of staining on WSP cards (not from an actual participant performance set). (A) Full WSP card with an example of staining. (B) Microscopy image of one salivary spot on a WSP card.

One tray (each tray consists of 7 WSP cards) was collected for each of three different “performance sets” per participant: (1) singing (if vocalist) or playing an instrument (if musician) without a mask/bell cover, (2) singing or playing an instrument with a mask/bell cover, and (3) talking without a mask. For 25 participants, an additional “blank” WSP card set was collected as a negative control. The blank WSP cards were taped to the magnetic inlet cone tray the same way as for the normal performance sets, but these trays were left in an open container placed on a table within the clean room approximately 4 feet away from the inlet cone for the duration of a participant’s time in the clean room.

Data Analyses

After each session, WSP cards were air-dried for 20 min and then stored in plastic zip-lock bags. Spots on the WSP cards were imaged with a Leitz Orthoplan microscope (85× magnification) fitted with a Progres Gryphax (NAOS 20.0MP CMOS Color, Feasterville, PA) digital camera. Under these settings, the smallest and largest spot sizes that could be resolved were 18 and 1770 μm in diameter, respectively. Each card was divided into 36 possible grid cells, and a random number generator selected 15 cells for imaging (105 images per set). Collected images were analyzed with ImageJ software (version 1.51, https://imagej.nih.gov/ij/all-notes.html) to quantify the area of each spot using methods described in Woolf et al.²⁹ Briefly, a color thresholding algorithm was developed in ImageJ to detect the edges of the blue spot stains against the yellow WSP card backgrounds. To optimize this algorithm, spot stains for 29 WSP cards were counted manually under a microscope, and RGB color thresholds were adjusted to minimize the mean absolute error between the number of spots counted manually and with the color thresholding algorithm. The final iteration of this algorithm produced a mean absolute error of ±3% for the test set. No adjustments were made for overlapping spots because the spot density was too low on each card for spot overlap to be a problem. See Figure S4 for an example of the results from this algorithm.

Data on spot count, size, and location were imported into R, version 4.0.2 (https://www.r-project.org/), for further analysis. A spread-factor equation

1

was used to convert spot diameter (d_s) to aerodynamic particle diameter (d_p) at the time of impact, following the calibration provided by Harrison et al.⁵ To estimate the original emitted particle diameters, we modeled the change in diameter between the point of release and WSP contact using a deterministic model (2.5 ms time steps) based on the rate of particle evaporation

2

and particle settling

3

where d_p is the particle aerodynamic diameter, D is the molecular diffusivity of water vapor in air at 293 K, M is the molecular weight of water, R is the universal gas constant, ρ_p is the density of liquid saliva,³⁰P_g is the vapor pressure of water at ambient conditions, T_g is the ambient temperature, P_d is the vapor pressure of water at the particle’s surface, T_d is the temperature at the particle’s surface, V_TS is the particle terminal settling velocity, g is the acceleration due to gravity, C_c is the Cunningham Correction factor, and η is the dynamic viscosity of air.³¹ Ambient RH was measured by an Omega Engineering HX92BC series sensor in the room at about 1.5 m from the inlet cone. A 6 cm travel distance between the WSP and the emission point was assumed for all particles. This distance was chosen based on observations of the typical distance between a participant’s mouth/instrument bell and the center of the WSP card tray. Participants were reminded to keep their instrument bells or mouths as close to the entrance plane of the inlet cone as possible at the beginning of each performance set (2–3 cm from the center of the WSP tray); however, this distance inevitably fluctuated during performances, and a 6 cm distance was observed as the average across all participant performance sets. Sensitivity analyses (Figures S5) indicated that a 6 cm (assumed) travel distance had minimal impact on size-resolved results, only producing a maximum error of ±4 μm for the smallest, most sensitive, size bin measured when varying the assumed travel distance between 3 and 9 cm. Particle counts were aggregated into log-equivalent size bins based on their aerodynamic diameters at their point of release and summed for each participant and card location. Counts were then blank-corrected by size and extrapolated to each card’s total surface area.

To estimate the full-plume emission rates, we interpolated the spatial distribution of summed particle counts, grouped by activity and size bin, between the 7 card locations and then extrapolated this pattern across our measurement domain using a bicubic spline extrapolation (R Akima::interp function; see the Supporting Information for additional information). The extrapolated area consisted of the lower half of the sampling inlet cone and extended approximately 17 cm downstream into the sampling duct. Figure S9 provides an example of the bicubic spline extrapolation output for the talking performance sets. Each facet in this figure represents a size bin, and the number in parentheses on top of each tile shows the extrapolation coefficient for that size bin. These coefficients were calculated by dividing the average number of particles estimated with the bicubic spline extrapolation by the average number of particles detected experimentally for each size bin and activity type and were then multiplied by the background corrected and surface-area-ratio-expanded counts for each individual performance set. Table S1 summarizes average experimentally detected counts versus the average bicubic spline extrapolated counts for each activity. These calculations resulted in estimates for the total count and size distribution of each participant’s performance set at the point of release. Particle emission rates are reported as total counts (i.e., total number of particles divided by the event duration, #·min^–1), size-resolved counts (#·min^–1 per size bin), and total mass (i.e., converting the size-resolved counts to mass, assuming particle density³⁰ of 1007 kg·m^–3 and taking the sum, μg·min^–1).

Linear mixed models were developed to examine associations between large-particle emissions and the following variables: instrument type (brass vs woodwind), participant age (adult vs minor), and sex (male vs female). The estimated variance for the participant-specific random effect was zero, indicating no correlation between an individual’s talking and instrument playing (or singing) emissions rates. Models that additionally adjusted for instrument or vocalization class (i.e., brass, woodwind, talking, or singing) were fit to assess pairwise differences between classes with Tukey adjustment. These models were developed only for the instruments (no bell cover), singing (no masks), and the talking performance sets due to small sample sizes in the other performance set types. Emission rates were log-transformed to meet the model assumptions. The models took the form:

where Y_ij is the log-transformed emission rate for the jth measurement from the ith participant, X_i is the set of fixed predictor variables, β is the vector of coefficients for the predictor variables, α_i is a random intercept term for participant i, and ϵ_ij represents the residual model error (assumed to have a mean of zero, be normally distributed, and have constant error variance).

We additionally assessed the ability of age, sex, CO₂ emissions, and loudness (dBA) to explain variation in large particle emissions by calculating the coefficient of determination (i.e., R²) from multiple linear regression models fit separately to the instrument and talking data sets.

Results and Discussion

Data were collected from 70 participants (37 adults and 33 minors). Of these, 31 (44%) were female (self-reported sex assigned at birth) and 39 were male (56%). A total of 168 performance sets were examined: 52 from talking, 11 from singing without a mask, 48 from instrument playing without a bell cover, 4 from singing with a mask, 28 from instrument playing with a bell cover, and 25 blank sets. A breakdown for all the performance set types is provided in Table S2. Descriptive statistics for count and mass emission rates (for nonmask/bell cover performance sets), as a function of performance set type and participant demographics, are shown in Table 1.

Table 1. Descriptive Statistics for Count Emission Rates (#·min^–1) and Mass Emission Rates (μg·min^–1) as a Function of Performance Set Type and Participant Demographics for Nonmask/Bell Cover Performance Sets.

	large particles per minute (#·min–1)
	brass	woodwinds	singing	talking	adults	minors	females	males
geometric mean	3.8	5.6	54.3	95.1	26.8	22.5	23.8	25.4
geometric SD	3.1	2.9	3.3	3.8	6.7	6.6	6.8	6.5
minimum	1.0	1.0	5.0	1.0	1.2	0.7	0.7	1.2
25%	1.5	2.4	33.3	57.2	6.6	4.2	3.7	6.0
median	3.4	4.0	93.0	93.0	38.9	26.8	39.1	25.5
75%	7.0	10.9	117.0	231.8	145.1	91.2	91.0	145.4
maximum	27.6	51.7	162.3	710.6	591.2	710.6	656.8	710.6

	large particle mass per minute (μg·min–1)
	brass	woodwinds	singing	talking	adults	minors	females	males
geometric mean	0.2	0.4	31.7	40.1	5.9	3.9	5.4	4.4
geometric SD	3.2	6.4	6.1	10.3	23.6	21.2	25.0	20.6
minimum	0	0	0	0	0.1	0	0	0.1
25%	0.1	0.2	26.1	11.3	0.4	0.2	0.2	0.3
median	0.2	0.3	64.7	41.9	11.2	2.7	12.8	3.7
75%	0.4	0.6	71.0	166.9	79.4	38.3	66.6	51.3
maximum	1.7	350.6	241.3	7507.2	7507.2	4007.3	7507.2	4007.3

Large particle emissions varied substantially among performers and within each type of performance set, consistent with prior studies, as shown in Figure 2.

Figure 2.

Boxplots of (A) count and (B) mass emission rates for large particles (d_p > 35 μm) from singing, talking, and playing wind instruments (for nonmask/bell cover performance sets). Center lines delineate the median; boxes represent the interquartile range, and whiskers represent 1.5·IQR or the data minimum. Each dot represents a participant.

For example, count emissions from talking ranged from as low as 1 min^–1 to as high as 711 min^–1. This large variation also provides evidence for the existence of super-emitters³² (i.e., people with emissions >1 geometric standard deviation beyond the geometric mean of the data) within the large particle size range (>35 μm), with the top 5 talking emitters (∼10% of participants) accounting for nearly 35% of total talking emissions. No obvious demographic patterns were found among the highest talking emitters, with 3 minors and 2 adults and 3 females and 2 males, though conclusions are limited with the small sample size (n = 5 for the top 10% of emitters). When examining if super-emitting tendencies for these individuals was consistent across performance activities, the data were inconclusive. Individuals who produced the highest large particle emission rates for talking were not the same individuals who produced the highest rates for their respective musical performance categories. Schlenczek et al. reported similar findings (albeit with a small sample size) in their study where the super-emitters for breathing were not the same as the super-emitters for instrument playing.⁹ Future studies should utilize many replicate samples from the same individuals to gather a more statistically robust data set to better understand if super-emitting tendencies for individuals are common across activity types and examine the temporal variability among these individuals.

Results show that, despite the large interparticipant variability, distinct differences were evident between the emission rates for the vocalization (singing and talking) and instrument playing (brass and woodwind instruments) categories, with vocalization producing on average 19.8 times (95% CI: 12.3, 31.8) higher emission rates than instrument playing (see Table S3 for model output and Figure 2 for graphical results). These results were similar to Schlenczek et al., who reported that wind instruments produce 30–600 times lower concentrations for d_p > 10 μm compared with talking and singing.⁹ Within the categories of vocalization and instrument playing, pairwise comparison testing indicated nonsignificant differences between talking and singing (p = 0.57) and between brass and woodwind instruments (p = 0.50); see Table S5 for model results. These nonsignificant differences run contrary to the previously reported results for smaller particle sizes quantified using the same participant performance sets,^26,27 where singing had significantly higher emission rates than talking and brass instruments had significantly higher rates than woodwind instruments. On the other hand, for large particles, our results align with Schlenczek et al., who reported similar emission rates of d_p > 50 μm particles between speaking and singing.⁹ We hypothesize that the discrepancy between large and small particles is the result of different generation mechanisms for these different size ranges.^22,33 Smaller respiratory particles (d_p < 10 μm), for example, are primarily generated via bronchial fluid film burst and vibration of the vocal folds, while larger particles (d_p > 35 μm) likely originate from movement of the lips and tongue during vocalization. We hypothesize two potential modes for large particle (d_p > 35 μm) emissions from wind instruments. The first is from saliva at and around the instrument mouthpiece and aerosolized due to “buzzing” by the player. The second is saliva accumulation within the instrument and subsequent droplet release due to vibrations of the instrument walls during playing. However, given the tortuous paths that air must follow to escape most wind instruments; this latter hypothesis seems less likely. These results when combined with those of Volckens et al.²⁷ suggest that, while wind instruments produce a lot of small particles, they produce very few large particles.

McCarthy et al.²¹ also examined large particle emissions from wind instruments using WSP from a panel of 9 participants. Unlike our results, they found no evidence of large particle emissions from instrument playing. Since McCarthy et al. utilized a smaller sample of participants and were testing for emissions caused only by playing single notes, it is possible that emissions were missed due to the nature of the experimental setup and size of the study. Schlenczek et al. used a holographic setup to detect large particles produced by wind instruments, and although they found no evidence of large particles emitted from the instrument bell, they detected a substantial amount of d_p > 35 μm particles emitted from the instrument mouthpiece, suggesting particle filtering by the instrument tubing as a main factor.⁹ This filtering of large particles on the inner tubing of the instrument is supported by Viala et al. who employed a computational fluid dynamics simulation for a clarinet and reported that particles > 50 μm will deposit inside the instrument before being emitted from the bell.¹⁰

From a mass standpoint, the differences in emission rates found in this study between performance set types within the categories (i.e., talking vs singing in the vocalization category) were also insignificant. However, the difference in emission rates between the vocalization and instrument playing categories was even larger than from a particle count basis, with vocalization mass emission rates on average exceeding those for playing instruments by a factor of 132 (95% CI: 60.5, 289) (Figure 2B and Table S4), while from a count basis, vocalization only exceeds emissions from instrument play by a factor of 19.8 (95% CI: 12.3, 31.8). This large difference between particle mass emission rates and particle count emission rates elucidates how vocalization performance sets produced a higher proportion of particles on the larger side of their size distribution compared to the instrument performance sets.

Count emission rates as a function of particle size and performance activity are listed in Figure 3.

Figure 3.

Large particle emission rate size distribution plot. The error bars for each size bin represent the standard error across all participants for the respective activity type.

Vocalization produced higher emission rates for every size bin, but especially toward the larger side of the size spectrum. Xie et al. measured large particle emissions produced by talking from a panel of 7 participants, using WSP cards and glass slides.²³ They determined that about 14% of count emissions came from the 40–50 μm size bin and about 28% came from the 50–75 μm size bin. Our study produced about 42% of count emissions coming from the 42–51 μm size bin and about 30% from the 51–74 μm bin. Johnson et al.²² conducted similar large particle emission quantification experiments for talking with 8 participants but produced quite different results, finding a substantial mode for large particle emissions at 145 μm, which we did not find in this study. Similarly, Harrison et al.⁵ also found no evidence for a secondary mode of large particle emissions around 145 μm. They also reported insignificant differences for count emissions between their talking and singing groups and showed large interparticipant variability between these groups. However, they reported emission rates over twice as high (median values of spanning 227–276 particles per minute among their talking, singing, and child adult cohorts) than what we found. This discrepancy may be due to differences in experimental setup and natural variation between study populations.

Because this study on large particle emissions was run concurrently to the Good et al.²⁶ study that focused on small particle emissions from vocalization (0.25 μm < d_p < 35 μm), there were 42 participants who provided talking data for both small and large particle emissions (spanning a size range from 0.25 to 1000 μm). Differences between the count and mass emission rates for these participants are listed in Figure 4.

Figure 4.

Comparison of particle emission rates between “Large” particles (d_p > 35 μm) and “Small” particles (d_p < 35 μm) for talking performance sets. (A) Count emission rates. (B) Mass emission rates.

While participants’ large particle count emission rates were much lower than those for their small particles (Figure 4A), their large particle mass emission rates were substantially higher than their small particle mass emission rates (Figure 4B). This difference is due to the cubic relationship between particle diameter and particle mass and is an important point to understand with relation to the spreading of communicable respiratory diseases, where fomite transmission can be a principal route for disease transmission. Like Harrison et al.,⁵ we found little correlation (R² = 0.072) between the large and small particle category count emission rate among the 42 participants who had talking data spanning this full size range, as shown in Figure S10.

An interesting finding from this work came from looking at how large particle emissions changed with an increased participant loudness and carbon dioxide (CO₂) output. Previous work^26,27,32 suggests CO₂ output and sound pressure level are correlated with increasing particle emissions for the smaller aerosolized size range (0.25–35 μm), and since larger particles are produced via similar physiological pathways, we hypothesized large particles would also have strong correlations with CO₂ and sound volume. However, we found no relationship between these predictors when assessing the strength of these correlations for the large particle emissions. When isolating for only the effect of sound pressure level (voice/instrument volume), no meaningful correlation was found with large particle emissions for any of the performance set types tested, with an R² ranging from 0.055 for talking to 0.067 for brass instruments. This is different from our prior work on the smaller particle sizes (0.25–35 μm) that suggested a statistically significant correlation between aerosol emission rates and sound pressure for brass instruments (R² = 0.357) and for vocalization (R² = 0.370).^26,27 In a similar fashion, we observed a very low correlation between large particle emission rates and average CO₂ concentrations in sampled air. The factors we measured did not explain variation in emissions rates well, with only 16.1% of variation in large particle emissions from instruments and 8.5% of variation from talking explained by age, sex, loudness, and CO₂. This suggests that variation in large particle emissions is dependent on variables we did not record for this study (i.e., variables beyond age, sex, instrument type, average CO₂, and average dBA). Variables of interest for future investigation include characterizing differences in individuals’ saliva production or variances in tongue and lip articulation.^25,34

The effect of bell covers on mitigating large particle emission rates from instruments is shown in Figure 5A; a similar relationship for the effects of masks on singing is shown in Figure 5B.

Figure 5.

Emission rates with and without mitigation. Points represent individual performance set emission rates and lines connecting two points indicate that the same individual produced both sets. (A) Count emission rates of 6 instruments with and without a bell cover. (B) Singing count emission rates with and without a mask.

These plots show the change in measured emission rate on a per-participant basis; the left side of each panel shows an individual’s count emission rate without a bell cover/mask, and the right side shows their emissions with a bell cover/mask. Lines connecting two dots indicate that these two performance sets were conducted by the same person. Most of the instruments tested had some people whose emission rates increased and some people whose rates decreased when using a bell cover. These results provide little evidence to suggest that bell covers either mitigate or enhance large particle emission rates for wind instruments (which are very low to begin with), counter to what was seen for the smaller aerosols²⁷ where the use of bell covers produced a statistically significant effect for brass instruments. The counterintuitive effect of the bell cover (Figure 5) could be due to the flow resistance imparted by the bell cover, which produced unintended consequences during playing. In other words, the additional pressure drop from the bell cover may require that participants blow harder into their mouthpiece, resulting in more particles being expelled from the instruments’ keyholes or from around participants’ mouths (i.e., increased “buzzing”); such an effect would likely vary from instrument to instrument, but the hypothesis remains untested. On the other hand, for the vocalization sets, wearing a mask proved to decrease large particle emissions substantially with an average decrease of 92.5% (95% CI: 97.9%, 73.7%). As shown in Figure 5B, all but one participant had dramatic reductions in large particle emissions (one participant already had a low emission rate without a mask).

This reduction in large particle emissions for singing while wearing a mask is further illustrated by the emission rate size distribution graph shown in Figure 6. Here, it is evident that wearing a mask eliminates all the largest size bins of particle emissions for singing and significantly reduced the emission rate for the smallest size bin we tested (42–51 μm). The substantial reduction in large particle emissions for singing while wearing a mask is consistent with the mask efficiency findings from previous studyies investigating smaller aerosols.^27,35,36 It should be noted that most of the particles considered in this study (d_p > 35 μm) would have sufficient inertia (whether by settling or being present in the jet of exhaled air) to be filtered out by masks via impaction.³¹ Although we did not evaluate the effects of masking on large particle emissions from talking, we cannot think of a credible reason wearing one would not also decrease this emission rate, given its similarity in mode of production, particle size distribution, and emission rate to that of singing.

Figure 6.

Large particle emission rate size distribution plot for singing with and without a mask.

Overall, humans emit fewer large salivary particles (35–1000 μm) than smaller respiratory particles (0.25–35 μm), an average of 415 (standard deviation of 1090) times less for talking. However, the mass of the larger particles far outweighs that of the smaller particles by an average factor of 30 300 (standard deviation of 147 350) when talking. Although these larger particles do not stay suspended in the air for as long as smaller particles, their increased volume implies that, on a per-particle basis, they may contain higher amounts of potentially pathogen-laden biological material and therefore pose a considerable risk for disease transmission, especially for respiratory diseases where large particles are the primary vector for transmission. Human vocalization (talking and singing) produces about 19.8 times more large particle emissions on a count basis and about 132 times more emissions on a mass basis than instrument playing (brass and woodwind). However, within these two categories, there is no statistically significant difference between the emissions rates for either talking vs singing or brass vs woodwind instrument playing. We also found through multiple linear regression modeling that age, sex, instrument type, CO₂ production, and sound pressure level do not explain a meaningful amount of variation for large particle emission rates for any performance category. Lastly, we determined that using a bell cover had negligible effects on large particle emission rates from the already low emissions detected for brass and woodwind instrument playing but that wearing a mask dramatically reduces large particle emission rates for singing.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.est.3c03588.

• Additional details on experimental setup, full emission plume extrapolation method, original particle size calculation, sensitivity analysis, and statistical modeling outputs (PDF)

The authors declare no competing financial interest.