Aerosol emission is increased in professional1
singing2
Dirk M¨urbe1,*,+, Martin Kriegel2,+, Julia Lange2, Hansj¨
org Rotheudt2, and Mario Fleischer1
3
1Charit´
e – Universit¨
atsmedizin Berlin, Department of Audiology and Phoniatrics, Berlin, 10117, Germany4
2Technische Universit¨
at Berlin, Hermann-Rietschel-Institut, Berlin, 10587, Germany5
*dirk.muerbe@charite.de6
+these authors contributed equally to this work7
ABSTRACT
8
In this study, emission rates of aerosols emitted by professional singers were measured with a laser
particle counter under cleanroom conditions. The emission rates during singing varied between 753.4
and 6095.37 P/s. Emission rates for singing were compared with data for breathing and speaking.
Significantly higher emission rates were found for singing. The growth rates between singing and
speaking were between 3.97 and 99.54. Further, effects of vocal loudness and gender were investigated.
The present study should support the efforts to improve the risk management in cases of possible
aerogenic virus transmission, especially for choir singing.
9
Introduction10
The respiratory system is the main transmission route for SARS-CoV-2-viruses1,2.11
Depending on particle size, a distinction can be made between droplets with a diameter greater than
12
5
μ
m and particles smaller than 5
μ
m (aerosols or droplet nuclei)
3–5
. Droplets and aerosols differ according
13
to the influence of gravity. For example, droplets of a size of 100
μ
m sink to the ground within a short
14
time and are transported up to a distance of 1.5 m6,7.15
When aerosols are exhaled, the fluid component of the pathogen-containing particles evaporates more
16
and more. They become lighter, can float in the air for longer periods and spread in closed rooms by
17
air flow and diffusion
8
. As the basis of a possible aerogenic transmission of the SARS-CoV-2-virus, the
18
spatial distribution of aerosols is dependent on several factors of the surrounding air, such as temperature
19
and humidity9.20
Droplets and aerosols are also produced during speaking and singing, because the respiratory tract
21
has a dual function: it is not only the main tool for ventilation, but also the source of voice and spoken
22
language production. Particle formation in the pulmonary alveoli
10
, flow effects of the vibrating vocal
23
folds and adjustments of the articulation instruments are regarded as aerosol generating mechanisms11.24
In comparison to breathing, a stronger formation of aerosols is known for speaking, whereby also
25
a dependence of the number of the arising particles on vocal loudness is described
12,13
. For singing,
26
a significantly higher aerosol production is assumed, probably due to the underlying physiological
27
mechanisms and the greater continuity of voice production over time. This assumption is supported by
28
reports of high infection rates during choir rehearsals in closed rooms14.29
Previous measurements focus on fluid mechanical aspects in the near-field plume of the mouth during
30
singing
6,15
. The spread of the emitted droplets is investigated, hence distance rules can be derived for
31
protection against droplet infection. However, a risk assessment including the distribution of aerosols in
32
larger rooms is not possible with this method.33
The current investigations aim to initially determine the number and size distribution of even small
34
particles emitted in the room by professional singers during singing. This information can be the basis
35
for a numerical calculation of the distribution of aerosols in larger rooms, which takes into account the
36
boundary conditions being typical for concert and opera performances.37
The present data may contribute to improved risk management strategies in the fields of culture and
38
education. They should be used for specification of hygiene measures and ventilation concepts in order to
39
facilitate performances and events.40
Results41
Particle size distribution42
The particle count measurement method detects different sizes of particles from 0.3
μ
m to 25
μ
m. As
43
shown in the log-probability plot (Fig. 1),
>
99 % of all detected particles were
≤
5
μ
m (
>
80 % of all
44
particles
≤
1
μ
m). Based on this observation, and following the agreement that aerosol particles of size
45
≤
5
μ
m are referred to as aerosol particles, the following results are given for particles of size 0.3
μ
m –
46
5μm.47
Experiment I48
Figure 2 illustrates both the emission rates for the different test conditions (breathing, speaking, and
49
singing) and the maximum sound pressure levels for singing.50
2/21
The results confirm the hypothesis of higher emission rates for singing compared to breathing and
51
speaking.52
While the individual median values for singing ranged from 753.36 P/s (S5) to 6095.37 P/s (S2)
53
(Table 1), those for speaking ranged from 14.13 P/s (S6) to 390.84 P/s (S1) (Table 1). The individual
54
median values for breathing ranged from 4.71 P/s (S1) to 428.55 (S2) (Table 1).55
The growth rate of the emission rates for singing in comparison to speaking was between 3.97 (S1)
56
and 99.54 (S2). Moreover, the growth rate of the emission rates for singing in comparison to breathing
57
was between 14.22 (S2) and 329.61 (S1) (Table 2).58
The evaluation of the sound pressure levels showed that the higher voice classifications soprano
59
(female) and tenor (male) had the expected higher sound pressure levels than the lower voice classifications
60
alto and baritone. While the maximum sound pressure level of males in the selected sample were always
61
positively correlated with the particle emission rate, there was no clear correlation in this respect for the
62
female voices.63
Statistical analysis by means of linear mixed modeling (Equation 2) showed significant differ-
64
ences of the (logarithmic) emission rate
log10PM
between the different test conditions breathing, speak-
65
ing and singing. Condition affected
log10PM
(
χ2
(2)=37.797, p=6.2
·
10
-9
) increasing it by a factor of
66
0.5230±0.2664
(standard errors) from breathing to speaking and by a factor of
1.7740±0.1211
(stan-
67
dard errors) from breathing to singing. By-subject analysis turned out that S2 and S6 showed a decrease of
68
emitted particles from breathing to speaking (see Fig. 2).69
Further, female singers showed significantly higher emission rates than males. Gender affected
log10PM
70
(
χ2
(1)=4.3035, p=0.03803) lowering it by a factor of
−0.3453±0.1246
(standard errors) from female to
71
male.72
Experiment II73
The results of measurements with the sustained vowel /a/ at different loudness conditions are presented in
74
Fig. 3. Seven of the eight subjects showed an increase in the emission rate with increasing loudness. The
75
comparison of piano (Table 3) and forte (Table 3) showed a growth rate up to 114.29 (S3) (Table 2). There
76
were higher emission rates for singing in forte for females (from 2023.02 P/s (S1) to 8072.35 P/s (S3))
77
compared to males (from 376.7 P/s (S5) to 2851.02 P/s (S7)). The same implications were made during
78
the increase from piano to mezzo-forte (see also Table 2).79
Statistical analysis by means of linear mixed modeling (Equation 2) showed significant differences of
80
3/21
the emission rate
log10PM
for the different vocal loudness conditions piano, mezzoforte and forte. Vocal
81
loudness affected
log10PM
(
χ2
(2)=12.47, p=0.00196) lowering it by a factor of
−0.45994±0.1119682
(standard errors) from forte to mezzoforte and by a factor of
−1.25514±0.23734
(standard errors) from
83
forte to piano. The described higher emission rates for females than for males failed to reach statistical
84
significance.85
For all subjects, the intended increase in loudness from piano to forte was reflected in the measured
86
values of the sound pressure level. Additionally, Figure 4 shows the relationship between the emission
87
rate and the maximum sound pressure level (only the median values for experiment II – sustained vowel
88
/a/ – were considered). An increase in the sound pressure level was accompanied by a mean increase in
89
the emission rate
log10PM
by a factor of 0.06. With regard to sustained vowels, it could be stated that the
90
emission rates can vary by more than two orders of magnitude.91
Discussion92
Due to the increased risk of transmission of SARS-CoV-2 viruses during singing and the described accu-
93
mulation of these infections during choir rehearsals, the survey of particle emissions and the assessment of
94
aerosols in rooms are key elements in the risk management of ensemble and choir singing in enclosed
95
rooms.96
The measuring method used (laser particle counter) provides very high accuracy concerning the
97
absolute number of particles and their size because sources of interference have been reduced to a
98
minimum. Furthermore, the suitability of the peripheral test setup could be proven within the scope of
99
baseline measurements.100
An alternative or supplemental method to investigate the size distribution of droplets during breathing,
101
speaking and singing is the imaging technique of Particle Image Velocimetry (PIV). This is based on
102
high-resolution photos of the particles, which are illuminated with a laser light, for example. Studies
103
using PIV also show that more particles are emitted when speaking loudly than speaking with low voice
15
.
104
However, mainly qualitative statements can be made here, due to several influencing factors. Size and
105
number of particles can only be estimated, because of the background concentration of particles in the
106
room and some drops can only be picked up in a blurred way. In a recent study
16
, particles of the sizes 1,
107
10, and 100
μ
m were measured with PIV and high accuracy was shown for particles greater than 6
μ
m.
108
This may be a reason why investigations of the size distribution of droplets with PIV lead to significantly
109
higher mean particle diameters
17
. Recent studies show that with PIV, particles in the order of 1
μ
m can
110
4/21
be examined
6
. For particles, in the order of 0.3 – 20
μ
m, the laser particle counter used in cleanroom
111
conditions offers higher accuracy in determining the number and size of particles.112
It should be noted, that the relative humidity of about 40%, the initial velocity at the mouth, and
113
constant airflow in the glass pipe lead to biased particle sizes as measured by the laser particle counter.114
According to Nicas et al.
18
, particles with an initial diameter of 20
μ
m shrink by a factor of 2, and
115
according to Wei & Li
7
particles up to 1000
μ
m shrink by a factor of 3 to the equilibrium diameter, nearly
116
independently of the relative humidity in the room.117
The size of this final state is dependent on the amount of non-soluble residues, on the humidity, and
118
the residual respiratory fluid. This final state, called droplet nuclei, is what is referred to as a particle in
119
the context of this article.120
In contrast, humidity has a highly relevant impact on the evaporation time of droplets. In general, the
121
evaporation time is proportional to the square of the initial diameter
7,8
. Considering the greatest measured
122
particle size of 5
μ
m in this study and a shrinking factor of 3, one gets droplets of a diameter of 15
μ
m in
123
maximum. Whereas a droplet of a size of 15
μ
m in diameter evaporates in dry air in about 0.15 s to its
124
resting state, the evaporation time increases at a relative humidity of 90% by a factor of about 25 to 3.75 s
125
(values were interpolated considering Table 1 in Wei & Li
7
). More than 80% of the particles measured in
126
this study are equal to or smaller than 1
μ
m (3
μ
m un-evaporated). For this particle size, evaporation times
127
are in the order of 0.006 s in dry air, and 0.08 s in moist air.128
Considering an initial fluid volume velocity at the mouth opening of about 0.142 (females) to 0.244
129
(males) liters per seconds for spoken vowels
19
, a lip opening area of about 300 to 1100 mm
2
for females
20
130
and 214.4 to 830.8 mm
2
for males
21
one gets fluid velocities in maximum in the order of magnitude of
131
0.47 (females) up to 1.14 m/s (males) for sustained vowels. These values are up to one order of magnitude
132
lower than reported values of 2.31–4.07 m/s
22
and 3.9 m/s
17
for speach. Adding these values to the
133
constant fluid flow velocity of 1.63 m/s in the glass pipe used in the experiment, droplets dispersed with a
134
velocity of 5.7 m/s in maximum over the length of 810 mm (Fig. 5).135
Taking into account the evaporation time of 3.75 s for 15
μ
m droplets, a characteristic evaporation
136
length of 21.375 m occurs. This length is lowered for droplets of the size of 3
μ
m to 0.456 m. Assumption
137
of dry air, these values drop considerably to 0.84 m for 15
μ
m droplets and 0.034 m for 1
μ
m droplets.
138
Recognizing a relative humidity of about 40% in the cleanroom, it can be expected that most of the emitted
139
droplets evaporates to droplet nuclei before reaching the laser particle counter. Additionally, because of
140
the small initial size of about 15
μ
m in the maximum of the particles, no dissection is expected and the
141
5/21
measured number of particles should be correct. To summarize these assessments, it should be noted
142
that the particle sizes measured are recognized as aerosols which distribute in an environment and not as143
particles that were emitted by a subject directly at the mouth. Thus, the droplet nuclei represent a realistic
144
measure for possible carrier particles for viruses.145
Since the aerosols emitted during breathing, speaking
12
and singing are mainly
<
1
μ
m in size, it
146
cannot be assumed that they sink quickly to the ground. It had been shown, that the retention time was in
147
the range of minutes to hours and the sink rate is in the order of
<
1 mm/s
4,7,8
. The determined order of
148
magnitude of the particle size of this study is significantly lower than the results of the only study, where
149
the particle emission during singing was also investigated. In this former study, the estimated particle
150
size during singing was determined with 68
μ
m in median
23,24
. Furthermore, in the same study, the
151
sizes of the emitted particles for speaking were determined by 81
μ
m. Apart from the methodological
152
aspect discussed above, the discrepancy between these and the data presented in this article, is probably
153
due to the high-precision measuring methods not yet available at that time. With regard to the size of
154
emitted particles, one was able to show that they are distinctively smaller than 10
μ
m during speaking and
155
breathing25,26.156
The present study confirms that higher emission rates of aerosols are produced during singing in
157
comparison to speaking and breathing. A higher emission rate for speaking compared to breathing and an
158
increase of emission rates with raising vocal loudness was found
25
. One could further show that the range
159
of emission rate ranges from 1 to 50 P/s for speaking
25
, which roughly confirms our data (14.13 to 390.84,
160
see Table 1). Furthermore, there is a good agreement in the emission rate in breathing25.161
However, phonation of sustained vowels, characterized by a periodic collision of the vocal folds
162
correlating with pitch, does not reflect the ordinary situation in choral singing. Here, the order of
163
consonants and vowels alternate in a sung passage and are interrupted by pauses. Therefore, in the present
164
study, a sequence of 50 seconds of the choir piece “Abschied vom Walde” by Felix Mendelssohn Bartholdy
165
was selected. Each line of the four-part choral movement was sung by the individually appropriate voice
166
classification (soprano, alto, tenor, baritone). These data were compared with the tasks ’breathing’ and
167
’speaking’ (reading the standardized text corpus). Again, there is an increase of the emission rate for
168
singing in comparison to speaking. Probably, this is due to the higher ratio of voiced segments to pauses
169
and the increased sound pressure level in singing. Further, these findings agree with the observation that
170
voiced vocalizations lead to higher aerosol emissions1,13.171
Apart from the influence of vocal loudness on the emission rate, we found gender differences with
172
6/21
higher emission rates for female singers. One reason for a stronger aerosol generation might be the
173
higher frequency of the vibrating vocal folds. This counts both, for the higher soprano and alto line of the
174
four-part choral movement and for the selected higher pitch for females during sustained phonation.175
However, the data presented here show no clear homogeneity within the cohort. For example, the
176
emission rate determined for singing fluctuates by almost one order of magnitude. Also, the increase
177
of
PM
between singing and speaking fluctuates by almost two orders of magnitude. Thus, the aspect of
178
high-emitters or super-emitters might be considered25.179
Of course, the determined emission rate does not provide any information about a possible concentra-
180
tion of SARS-CoV-2 viruses yet. However, at present this number can not serve to estimate the probability
181
of infection27.182
It should be noted that in the course of the actual pandemic so far, numerous situations seem to be
183
related to a high probability of aerogenic virus transmission, among them choir rehearsals. There is also
184
initial evidence of viable SARS-CoV-2 viruses in indoor air
28
. However, comprehensive information on
185
the transmission quantity and survivability of SARS-CoV-2 viruses in aerosols is still missing29.186
Therefore, the present study contributes to one component in the risk assessment of singing, which in
187
turn is largely determined by the current prevalence. Finally, there is a lack of data on whether specific
188
breathing characteristics of singing (deep inhalation, higher intrapulmonary pressures) influence the risk
189
of transmission when singing loudly. In any case, the data should support all efforts to improve the risk
190
management, especially in choir singing.191
Methods192
Subjects193
Eight singers (ages 22 to 62 years; professional choir experience between 1 to 34 years) of a professional
194
chamber choir (RIAS Kammerchor Berlin) took part in the investigations. To each of the different voice
195
classifications belonged two of the subject group: alto (S1 & S2), soprano (S3 & S4), baritone (S5 & S6),
196
and tenor (S7 & S8). This study was conducted according to the ethical principles based on the WMA
197
Declaration of Helsinki and to the current legal provisions and informed consent was obtained from all
198
subjects. It should be noted, that the results for breathing and speaking tasks of the subjects considered in
199
this study, have already been analyzed and published within a larger cohort
12
. In order to allow a direct
200
comparison with the data for singing, the data of this subgroup were reused and analyzed.201
7/21
Particle measurements202
The investigations were carried out in a cleanroom at the Hermann Rietschel Institute of the Technical
203
University of Berlin.204
The supply air was introduced via a vertical low-turbulence displacement flow (TAV) over the entire
205
ceiling area of 4.8 x 4.8 m
2
. The supply air velocity was 0.3 m/s and thus prevented thermal lift at the
206
people. The exhaust air was also discharged from the room over the entire surface via a raised floor. The
207
room temperature was
295.15 K±0.50 K
, the relative humidity was
40 %±2 %
and the room had 15 Pa
208
overpressure to the surrounding rooms.209
The actual test stand was located in this highly pure environment (Fig. 5). It consisted of a glass
210
pipe, in which a constant airflow of 400 m
3
/h was generated by a filter fan unit (Ziehl-Abegg, Künzelsau,
211
Deutschland). The measuring probe of a laser particle counter (Lighthouse Solair 3100 E, Lighthouse
212
Worldwide Solutions, Fremont, CA) was placed centrally in the pipe.213
The particle counter was counting with a volume flow
˙
VPC
of 28.3 l/min, with a measuring time of 10
214
seconds each and detected particles in six size classes:
>
0.3
μ
m – 0.5
μ
m,
>
0.5
μ
m – 1.0
μ
m,
>
1.0
μ
m –
215
3.0, >3.0 μm – 5.0 μm, >5.0 μm – 10 μm and >10 μm.216
The emission rate
PM
presented in Figs. 2–4was computed based on the measured particle concentra-
217
tion cMand the volume flow through the filter fan unit (FFU) ˙
VFFU , i.e.218
PM=cM·˙
VFFU .(1)
To estimate sources of interference, such as background noise of particles in the room, as well as
219
abrasion on the clothing and hair of the persons investigated, a baseline measurement was carried out at
220
the beginning of the investigation. For particle reduction due to movement artifacts, the test persons wore
221
cleanroom clothing and a headgear with the sealing of the edges with adhesive tape, so that only eyes,
222
nose, and mouth were uncovered.223
In this baseline measurement, a count rate of the particle counter of
<
1 particles/5 minutes was
224
determined within a measurement period of 10 minutes.225
The counting efficiency for particles of the size 0.3
μ
m is
50 %±20 %
and for particles of the size
226
0.5
μ
m it is
100 %±10 %
according to ISO 21501-4. To investigate how many particles were separated
227
over the measuring distance, comparative measurements were made over a short distance from the particle
228
counter. For this case, the particles were directly collected through a 150 mm high funnel while breathing
229
8/21
and speaking and directed to the particle counter. The same size distribution was found as in the finally
230
used configuration.231
Audio measurements232
The sound pressure level was determined using a calibrated sound level meter (CENTER 322_ Datalogger
233
Sound Level Meter, Center Technologies, Houston, TX). During all measurements, the sound level meter
234
was located approximately 60 cm anterior-laterally away from the mouth of the test persons due to limited
235
accessibility. The measuring arrangement of the particle counter did not allow a standard positioning of
236
30 cm mouth distance of the measuring device. Furthermore, the high sensitivity of the particle counter did
237
not allow a frontal positioning of the sound level meter inside the glass tube. Consequently, the determined
238
levels were not to be considered as absolute levels but are lowered by a constant value of approx. 10 dB
239
SPL.240
Due to the time variability of the determined sound pressure levels (primarily for speaking and singing),
241
the maximum value
LAFMAX
of the frequency- and time-weighted acoustic pressure was recorded and
242
evaluated.243
Test conditions244
The subjects were in a sitting position at the entry of the particle measurement setup. Two experiments
245
were carried out:246
Experiment I: Comparison of three different test conditions247
a) Breathing through the mouth248
b) Reading a standardized text249
c) Singing a line of a four-part choral movement250
Experiment II: Singing a sustained vowel (/a/) at three loudness conditions251
a) piano252
b) mezzo-forte253
c) forte254
9/21
For experiment I, respectively, a time window of 50 seconds was analyzed. Further, for experiment II
255
the time window was set to 10 seconds. For reading in a comfortable loudness condition (Ib), the text
256
“Der Nordwind und die Sonne” by Äsop was selected. To pass Ic) the choral part of the song “Abschied
257
vom Walde” by Felix Mendelssohn-Bartholdy was chosen. The subjects were instructed to sing the line of
258
their individual voice classification. Each of all tasks were repeated five times.259
The following pitches were selected for experiment II: soprano: C5 (523 Hz), alto: F4 (349 Hz), tenor:
260
C4 (262 Hz), and baritone: F3 (175 Hz). The total measuring time for all tasks was about 30 minutes for
261
each subject.262
Statistical Analysis263
Besides the description of the data, a confirmative analysis was carried out. Therefore, a linear mixed
effects analysis of the relationship between
log10PM
, gender, condition and subject was performed by
means of the freely available software package R
30
including the package lme4
31,32
. The model used was
log10PM∼Condition+Gender+(1+Condition|Subject)+(1+Gender|Subject)(2)
(R model syntax). Condition and gender were incorporated as fixed effects into the model. Intercepts for
264
subject were incorporated as random effects. To keep the model maximal
33
, by--subject random slopes
265
for the effect of gender and condition were additionally incorporated as random effects. The interaction
266
term between condition and gender was identified as not significant and therefore not regarded. Careful
267
visual inspection of residual-plots and Q-Q-plots did reveal obvious deviations from homoscedasticity and
268
normality. Therefore, log-transform of
PM
was considered which overcomes these problems. To avoid
269
infinite values in the analyses, only
PM>0
were taken into account. To test significance, the P-values
270
were obtained by likelihood ratio tests of the full model with the effect in question against the model
271
without the effect in question. For this reason, linear mixed models were fit by maximum likelihood to
272
enable comparison.273
References274
1.
Asadi, S., Bouvier, N., Wexler, A. S. & Ristenpart, W. D. The coronavirus pandemic and aerosols:
275
Does covid-19 transmit via expiratory particles? Aerosol Sci. Technol.
54
, 635–638, DOI: 10.1080/
276
02786826.2020.1749229 (2020). https://doi.org/10.1080/02786826.2020.1749229.277
10/21
2.
Morawska, L. & Cao, J. Airborne transmission of sars-cov-2: The world should face the reality.
278
Environ. Int. 105730, DOI: 10.1016/j.envint.2020.105730 (2020).279
3.
Couch, R. B., Cate, T. R., Douglas, R. G., Gerone, P. J. & Knight, V. Effect of route of inoculation on
280
experimental respiratory viral disease in volunteers and evidence for airborne transmission. Microbiol.
281
Mol. Biol. Rev. 30, 517–529 (1966). https://mmbr.asm.org/content/30/3/517.full.pdf.282
4.
Tellier, R. Review of aerosol transmission of influenza a virus. Emerg. infectious diseases
12
,
283
1657–1662 (2006).284
5.
Judson, S. D. & Munster, V. J. Nosocomial transmission of emerging viruses via aerosol-generating
285
medical procedures. Viruses 11, 940 (2019).286
6.
Kähler, C. J. & Hain, R. Fundamental protective mechanisms of face masks against droplet infections.
287
J. Aerosol Sci. 148, 105617, DOI: https://doi.org/10.1016/j.jaerosci.2020.105617 (2020).288
7.
Wei, J. & Li, Y. Enhanced spread of expiratory droplets by turbulence in a cough jet. Build. Environ.
289
93, 86 – 96, DOI: https://doi.org/10.1016/j.buildenv.2015.06.018 (2015).290
8.
Stadnytskyi, V., Bax, C. E., Bax, A. & Anfinrud, P. The airborne lifetime of small speech droplets and
291
their potential importance in sars-cov-2 transmission. Proc. Natl. Acad. Sci.
117
, 11875–11877, DOI:
292
10.1073/pnas.2006874117 (2020). https://www.pnas.org/content/117/22/11875.full.pdf.293
9.
Morawska, L. Droplet fate in indoor environments, or can we prevent the spread of infection? Indoor
294
Air
16
, 335–347, DOI: 10.1111/j.1600-0668.2006.00432.x (2006). https://onlinelibrary.wiley.com/
295
doi/pdf/10.1111/j.1600-0668.2006.00432.x.296
10.
Johnson, G. R. & Morawska, L. The mechanism of breath aerosol formation. J. Aerosol Medicine
297
Pulm. Drug Deliv. 22, 229–237, DOI: 10.1089/jamp.2008.0720 (2009).298
11.
Johnson, G. et al. Modality of human expired aerosol size distributions. J. Aerosol Sci.
42
, 839–851,
299
DOI: 10.1016/j.jaerosci.2011.07.009 (2011).300
12.
Hartmann, A., Lange, J., Rotheudt, H. & Kriegel, M. Emission rate and particle size of bioaerosols
301
during breathing, speaking and coughing (2020). DOI: 10.14279/depositonce-10331. Preprint
302
Technische Universität Berlin.303
13.
Asadi, S. et al. Effect of voicing and articulation manner on aerosol particle emission during human
304
speech. PLOS ONE 15, 1–15, DOI: 10.1371/journal.pone.0227699 (2020).305
11/21
14.
Hamner, L. et al. High sars-cov-2 attack rate following exposure at a choir practice – skagit county,
306
washington, march 2020. MMWR Morb Mortal Wkly Rep
69
, 606–610, DOI: 10.15585/mmwr.
307
mm6919e6 (2020).308
15.
Anfinrud, P., Stadnytskyi, V., Bax, C. E. & Bax, A. Visualizing speech-generated oral fluid droplets
309
with laser light scattering. New Engl. J. Medicine DOI: 10.1056/NEJMc2007800 (2020). https:
310
//www.nejm.org/doi/pdf/10.1056/NEJMc2007800.311
16.
Chen-Yu, C., Atkinson Joseph, F., VanBenschoten John, E., Bursik Marcus, I. & DePinto Joseph,
312
V. Image-based system for particle counting and sizing. J. Environ. Eng.
126
, 258–266, DOI:
313
10.1061/(asce)0733-9372(2000)126:3(258) (2000).314
17.
Chao, C. et al. Characterization of expiration air jets and droplet size distributions immediately at the
315
mouth opening. J. Aerosol Sci.
40
, 122 – 133, DOI: https://doi.org/10.1016/j.jaerosci.2008.10.003
316
(2009).317
18.
Nicas, M., Nazaroff, W. W. & Hubbard, A. Toward understanding the risk of secondary airborne
318
infection: emission of respirable pathogens. J. occupational environmental hygiene
2
, 143–154
319
(2005).320
19.
Jiang, J. J., Hanna, R. B., Willey, M. V. & Rieves, A. The measurement of airflow using singing
321
helmet that allows free movement of the jaw. J. Voice 30, 641–648 (2016).322
20.
Garnier, M., Henrich, N., Smith, J. & Wolfe, J. Vocal tract adjustments in the high soprano range.
323
The J. Acoust. Soc. Am. 127, 3771–3780 (2010).324
21.
Fleischer, M., Pinkert, S., Mattheus, W., Mainka, A. & Mürbe, D. Formant frequencies and bandwidths
325
of the vocal tract transfer function are affected by the mechanical impedance of the vocal tract wall.
326
Biomech Mod Mechanobiol 14, 719–733 (2015).327
22.
Kwon, S.-B. et al. Study on the initial velocity distribution of exhaled air from coughing and speaking.
328
Chemosphere 87, 1260–1264 (2012).329
23.
Loudon, R. G. & Roberts, R. M. Relation between the airborne diameters of respiratory droplets and
330
the diameter of the stains left after recovery. Nature 213, 95–96 (1967).331
24.
Loudon, R. G. & Roberts, R. M. Singing and the dissemination of tuberculosis. Am. Rev. Respir. Dis.
332
98
, 297–300, DOI: 10.1164/arrd.1968.98.2.297 (1968). PMID: 5667756, https://www.atsjournals.org/
333
doi/pdf/10.1164/arrd.1968.98.2.297.334
12/21
25.
Asadi, S. et al. Aerosol emission and superemission during human speech increase with voice
335
loudness. Sci. Reports 9, 2348– (2019).336
26.
Papineni, R. S. & Rosenthal, F. S. The size distribution of droplets in the exhaled breath of healthy
337
human subjects. J. Aerosol Medicine
10
, 105–116, DOI: 10.1089/jam.1997.10.105 (1997). PMID:
338
10168531, https://doi.org/10.1089/jam.1997.10.105.339
27.
Bar-On, Y. M., Flamholz, A., Phillips, R. & Milo, R. Science forum: Sars-cov-2 (covid-19) by the
340
numbers. Elife 9, e57309, DOI: 10.7554/eLife.57309 (2020).341
28.
Guo, Z.-D. et al. Aerosol and surface distribution of severe acute respiratory syndrome coronavirus 2
342
in hospital wards, wuhan, china, 2020. Emerg. Infect. Dis. journal 26, 1583 (2020).343
29.
van Doremalen, N. et al. Aerosol and surface stability of sars-cov-2 as compared with sars-cov-1.
344
New Engl. J. Medicine
382
, 1564–1567, DOI: 10.1056/NEJMc2004973 (2020). https://doi.org/10.
345
1056/NEJMc2004973.346
30.
R Core Team. R: A Language and Environment for Statistical Computing. R Foundation for Statistical
347
Computing, Vienna, Austria (2020).348
31.
Bates, D., Mächler, M., Bolker, B. & Walker, S. Fitting linear mixed-effects models using lme4. J.
349
Stat. Softw. 67, 1–48, DOI: 10.18637/jss.v067.i01 (2015).350
32.
Winter, B. Linear models and linear mixed effects models in r with linguistic applications (2013).
351
1308.5499.352
33.
Barr, D. J., Levy, R., Scheepers, C. & Tily, H. J. Random effects structure in mixed-effects models:
353
Keep it maximal. J. Mem. Lang. 68, 255–278 (2013).354
Acknowledgements355
We thank the members of the RIAS Kammerchor Berlin for their support.356
Author contributions statement357
D. M., M. F., and M. K. designed research. J. L., H. R. and M. F. made measurements. M. F., J. L., D. M.
358
and M. K. wrote the paper.359
13/21
40 50 60 70 80 90 100
Percentage of
PM
(sum over all subjects)
0.3
0.5
1.0
3.0
5.0
10.0
Particle size in m
breathing
speaking
singing
sustained phonation (piano)
sustained phonation (mezzoforte)
sustained phonation (forte)
Figure 1. Log-probability plot of the frequency distribution of the size of the detected particles.
Regardless of the task, >99 % of all detected particles are ≤5μm (dashed line). Furthermore all tasks
show that >80 % of all particles are ≤1μm.
15/21
Alto
S1 Alto
S2 Soprano
S3 Soprano
S4 Baritone
S5 Baritone
S6 Tenor
S7 Tenor
S8
100
101
102
103
104
105
PM
in P/s
breathing
speaking
singing
breathing
speaking
singing
breathing
speaking
singing
breathing
speaking
singing
breathing
speaking
singing
breathing
speaking
singing
breathing
speaking
singing
breathing
speaking
singing
40
20
0
20
40
60
80
100
LAFMAX
in dB SPL
Figure 2. Boxplots of the emission rates (bars represent the median) for different gender, voice
classifications and the test conditions breathing, speaking and singing in experiment I (left y-axis). Only
particles ≤5μm were considered. For singing, the maximum sound pressure levels LAFMAX are also
shown (full circles, right y-axis).
16/21
Alto
S1 Alto
S2 Soprano
S3 Soprano
S4 Baritone
S5 Baritone
S6 Tenor
S7 Tenor
S8
100
101
102
103
104
105
PM
in P/s
piano
mezzoforte
forte
piano
mezzoforte
forte
piano
mezzoforte
forte
piano
mezzoforte
forte
piano
mezzoforte
forte
piano
mezzoforte
forte
piano
mezzoforte
forte
piano
mezzoforte
forte
40
20
0
20
40
60
80
100
LAFMAX
in dB SPL
Figure 3. Boxplots of the emission rates (bar represents the median) for different gender, voice
classifications and vocal loudness conditions while sustaining the vowel /a/ (Experiment II) (left y-axis).
Only particles ≤5μm were considered. For the different loudness conditions, the maximum sound
pressure levels LAFMAX are also shown (full circles, right y-axis).
17/21
50 60 70 80 90 100
LAFMAX
in dB SPL
10 1
100
101
102
103
104
105
106
PM
in P/s
log
10
(
PM
) = 0.06
LAFMAX
-1.52 (R
2
=0.616)
95% Prediction band
95% Confidence region
female (sustained phonation only)
male (sustained phonation only)
Noise
Figure 4. Relationship between emission rate and the maximum sound pressure level for the test
condition of sustained vowel /a/ (Experiment II) for all three loudness conditions separated by gender
including linear regression of the logarithmic emission rates (black line). Only particles ≤5μm were
considered. The grey field represents the sound pressure level resulting from the environmental conditions
(primarily particle counter) alone.
18/21
2
filter ceiling
exhaust air perforated floor
baffle
faceplate
810
295
particle
counter
˙
VFFU
1
Figure 5. Left: Schematic test setup with one person in cleanroom clothing whose exhaled air was
recorded by the particle counter. The glass measuring section was located on the suction side of a
horizontally positioned Filter Fan Unit (FFU). All geometric dimensions are in mm (Figure adapted from
Fig 2 in Hartmann et al.12).
19/21
Table 1. Minimum, maximum, and median values of emission rates in P/s for breathing, speaking, and
singing
breathing speaking singing
ID Min Median Max Min Median Max Min Median Max
S1 4.71 4.71 310.46 136.46 390.84 677.64 946.24 1552.39 2666.86
S2 84.72 428.55 508.16 14.13 61.24 84.72 5370.32 6095.37 7177.94
S3 14.13 14.13 18.84 32.96 61.24 79.98 1399.59 1761.98 2009.09
S4 4.71 28.25 183.65 164.82 296.48 570.16 1256.03 1761.98 1954.34
S5 9.42 11.53 28.25 42.36 70.63 84.72 630.96 753.36 997.70
S6 4.71 56.49 457.09 4.71 14.13 37.67 734.51 860.99 970.51
S7 4.71 12.47 131.83 28.25 47.10 84.72 881.05 1078.95 1253.14
S8 4.71 6.67 32.96 122.46 127.06 805.38 941.89 1520.55 1694.34
Table 2. Ratios of medians of emission rates for different test and loudness conditions
ID Speaking/breathing Singing/breathing Singing/speaking Forte/piano Forte/mezzoforte
S1 82.99 329.61 3.97 60.81 6.61
S2 0.14 14.22 99.54 19.86 1.46
S3 4.34 124.74 28.77 114.29 2.50
S4 10.50 62.37 5.94 41.98 6.00
S5 6.12 65.31 10.67 0.84 1.46
S6 0.25 15.24 60.95 3.40 1.55
S7 3.78 86.50 22.91 40.36 1.59
S8 19.05 228.03 11.97 44.46 6.35
20/21
Table 3. Minimum, maximum, and median values of emission rates in P/s for piano, mezzoforte, and
forte
piano mezzoforte forte
ID Min Median Max Min Median Max Min Median Max
S1 23.55 33.27 494.31 235.50 306.20 588.84 1389.95 2023.02 2710.19
S2 23.55 188.36 353.18 1106.62 2564.48 3365.12 2094.11 3741.11 8609.94
S3 47.10 70.63 117.76 1294.20 3228.49 5248.07 6729.77 8072.35 10303.86
S4 23.55 94.19 353.18 588.84 659.17 729.46 3605.79 3953.67 5176.07
S5 353.18 447.71 517.61 211.84 258.82 376.70 164.82 376.70 824.14
S6 141.25 235.50 753.36 399.94 517.61 776.25 588.84 799.83 1294.20
S7 23.55 70.63 918.33 1531.09 1790.61 3061.96 2238.72 2851.02 5714.79
S8 23.55 47.10 376.70 94.19 329.61 588.84 1648.16 2094.11 3622.43
21/21
