facemaskefficacyforfilteringexpelleddroplets.pdf

Cite as: Fischer et al., Sci. Adv.
10.1126/sciadv.abd3083 (2020).

RESEARCH ARTICLES

First release: 7 August 2020 www.advances.sciencemag.org (Page numbers not final at time of first release) 1

Introduction
The global spread of COVID-19 in early 2020 has significantly
increased the demand for face masks around the world, while
stimulating research about their efficacy. Here we adapt a re-
cently demonstrated optical imaging approach (1, 2) and
highlight stark differences in the effectiveness of different
masks and mask alternatives to suppress the spread of res-
piratory droplets during regular speech.

In general, the term ‘face mask’ governs a wide range of
protective equipment with the primary function of reducing
the transmission of particles or droplets. The most common
application in modern medicine is to provide protection to
the wearer (e.g., first responders), but surgical face masks
were originally introduced to protect surrounding persons
from the wearer, such as protecting patients with open
wounds against infectious agents from the surgical team (3),
or the persons surrounding a tuberculosis patient from con-
tracting the disease via airborne droplets (4). This latter role
has been embraced by multiple governments and regulatory
agencies (5), since COVID-19 patients can be asymptomatic
but contagious for many days (6). The premise of protection
from infected persons wearing a mask is simple: wearing a
face mask will reduce the spread of respiratory droplets con-
taining viruses. In fact, recent studies suggest that wearing
face masks reduces the spread of COVID-19 on a population
level, and consequently blunts the growth of the epidemic
curve (7, 8). Still, determining mask efficacy is a complex
topic that is still an active field of research (see for example

(9)), made even more complicated because the infection path-
ways for COVID-19 are not yet fully understood and are com-
plicated by many factors such as the route of transmission,
correct fit and usage of masks, and environmental variables.
From a public policy perspective, shortages in supply for sur-
gical face masks and N95 respirators, as well as concerns
about their side effects and the discomfort of prolonged use
(10), have led to public use of a variety of solutions which are
generally less restrictive (such as homemade cotton masks or
bandanas), but usually of unknown efficacy. While some tex-
tiles used for mask fabrication have been characterized (11),
the performance of actual masks in a practical setting needs
to be considered. The work we report here describes a meas-
urement method that can be used to improve evaluation in
to guide mask selection and purchase decisions.

A schematic and demonstration image are shown in Fig.
1. In brief, an operator wears a face mask and speaks into the
direction of an expanded laser beam inside a dark enclosure.
Droplets that propagate through the laser beam scatter light,
which is recorded with a cell phone camera. A simple com-
puter algorithm is used to count the droplets in the video.
The required hardware for these measurements is commonly
available; suitable lasers and optical components are accessi-
ble in hundreds of research laboratories or can be purchased
for less than $200, and a standard cell phone camera can
serve as a recording device. The experimental setup is simple
and can easily be built and operated by non-experts.

Below we describe the measurement method and

Low-cost measurement of facemask eff icacy for filtering
expelled droplets during speech
Emma P. Fischer1, Martin C. Fischer2,3,*, David Grass2, Isaac Henrion4, Warren S. Warren2,3,5,6, and Eric
Westman7
1Department of Psychology & Neuroscience, Duke University, Durham, NC 27708, USA. 2Department of Chemistry, Duke University, Durham, NC 27708, USA. 3Department
of Physics, Duke University, Durham, NC 27708, USA. 4Cover Durham, Durham, NC 27701, USA. 5Department of Radiology, Duke University School of Medicine, Durham, NC
27710, USA. 6Department of Biomedical Engineering, Duke University, Durham, NC 27708, USA. 7Department of Medicine, Duke University School of Medicine, Durham, NC
27708, USA.

*Corresponding author. Email: [email protected]

Mandates for mask use in public during the recent COVID-19 pandemic, worsened by global shortage of
commercial supplies, have led to widespread use of homemade masks and mask alternatives. It is assumed
that wearing such masks reduces the likelihood for an infected person to spread the disease, but many of
these mask designs have not been tested in practice. We have demonstrated a simple optical measurement
method to evaluate the efficacy of masks to reduce the transmission of respiratory droplets during regular
speech. In proof-of-principle studies, we compared a variety of commonly available mask types and
observed that some mask types approach the performance of standard surgical masks, while some mask
alternatives, such as neck fleece or bandanas, offer very little protection. Our measurement setup is
inexpensive and can be built and operated by non-experts, allowing for rapid evaluation of mask
performance during speech, sneezing, or coughing.

Science Advances Publish Ahead of Print, published on August 7, 2020 as doi:10.1126/sciadv.abd3083

Copyright 2020 by American Association for the Advancement of Science.

o
n
A

u
g
u
st 2

7
, 2

0
2
0

h
ttp

://a
d
va

n
ce

s.scie
n
ce

m
a
g
.o

rg
/

D
o
w

n
lo

a
d
e
d
fro

m

http://www.advances.sciencemag.org/

http://advances.sciencemag.org/

Amy
Highlight

Amy
Highlight

First release: 7 August 2020 www.advances.sciencemag.org (Page numbers not final at time of first release) 2

demonstrate its capabilities for mask testing. In this applica-
tion, we do not attempt a comprehensive survey of all possi-
ble mask designs or a systematic study of all use cases. We
merely demonstrated our method on a variety of commonly
available masks and mask alternatives with one speaker, and
a subset of these masks were tested with four speakers. Even
from these limited demonstration studies, important general
characteristics can be extracted by performing a relative com-
parison between different face masks and their transmission
of droplets.

Results
We tested 14 commonly available masks or masks alterna-
tives, one patch of mask material, and a professionally fit-
tested N95 mask (see Fig. 2 and Table 1 for details). For ref-
erence, we recorded control trials where the speaker wore no
protective mask or covering. Each test was performed with
the same protocol. The camera was used to record a video of
approximately 40 s length to record droplets emitted while
speaking. The first 10 s of the video serve as baseline. In the
next 10 s, the mask wearer repeated the sentence “Stay
healthy, people” five times (speech), after which the camera
kept recording for an additional 20 s (observation). For each
mask and for the control trial, this protocol was repeated 10
times. We used a computer algorithm (see Materials and
Methods) to count the number of particles within each video.

The results of our mask study are depicted in Fig. 3 (A),
where we show the relative droplet count for each tested
mask. The data displayed with solid dots represent the out-
come of the same speaker testing all masks; the points and
error bars represent the mean value and distribution stand-
ard deviation, respectively, of the total droplet count normal-
ized to the control trial (no mask). For this speaker’s control
trial, the absolute droplet count was about 960. A graph with
corresponding logarithmic scale can be found in Supplemen-
tary Fig. S1. The data in Fig. 3 (A) displayed with a hollow
circle represents an average over four different speakers
wearing the same type of masks (surgical, cotton5, and ban-
dana); the values and error bars represent the mean value
and standard deviation of the average relative droplet count
from all four speakers. The additional speakers’ reference
counts for the control trial (no mask) were about 200, with
similar fractional variance to the main speaker (see Supple-
mentary Fig. S2 for details).

We measured a droplet transmission fraction ranging
from below 0.1% (fitted N95 mask) to 110% (fleece mask, see
discussion below) relative to the control trials. In Fig. 3 (B),
the time evolution of detected droplets is shown for four rep-
resentative examples (surgical, cotton5, bandana, and the
control trial) tested by the first speaker – the data for all
tested masks is shown in Supplementary Fig. S3. The solid
curves indicate the droplet transmission rate over time. For

the control trial (green curve), the five distinct peaks corre-
spond to the five repetitions of the operator speaking. In the
case of speaking through a mask, there is a physical barrier,
which results in a reduction of transmitted droplets and a
significant delay between speaking and detecting particles. In
effect, the mask acts as a temporal low pass filter, smoothens
the droplet rate over time, and reduces the overall transmis-
sion. For the bandana (red curve), the droplet rate is merely
reduced by a factor of two and the repetitions of the speech
are still noticeable. The effect of the cotton mask (orange
curve) is much stronger. The speech pattern is no longer rec-
ognizable and most of the droplets, compared to the control
trial, are suppressed. The curve for the surgical mask is not
visible on this scale. The shaded areas for all curves display
the cumulative particle count over time: the lower the curve,
the more droplets are blocked by the mask. Figure 3 (B)
shows the droplet count for the four masks measured by one
speaker; Supplementary Fig. S4 shows the data for all four
speakers using identical masks.

We noticed that speaking through some masks (particu-
larly the neck fleece) seemed to disperse the largest droplets
into a multitude of smaller droplets (see Supplementary Fig.
S5), which explains the apparent increase in droplet count
relative to no mask in that case. Considering that smaller par-
ticles are airborne longer than large droplets (larger droplets
sink faster), the use of such a mask might be counterproduc-
tive. Furthermore, the performance of the valved N95 mask
is likely affected by the exhalation valve, which opens for
strong outwards airflow. While the valve does not compro-
mise the protection of the wearer, it can decrease protection
of persons surrounding the wearer. In comparison, the per-
formance of the fitted, non-valved N95 mask was far superior.

Discussion
The experimental setup is very straightforward to implement,
and the required hardware and software are ubiquitous or
easily acquired. However, this simplicity does go along with
some limitations that are discussed here, along with routes
for possible improvements and future studies. Again, we
want to note that the mask tests performed here (one speaker
for all masks and four speakers for selected masks) should
serve only as a demonstration. Inter-subject variations are to
be expected, for example due to difference in physiology,
mask fit, head position, speech pattern, and such.

A first limitation is that our experimental implementation
samples only a small part of the enclosure and hence some
droplets that are transmitted through the masks might not
be registered in the laser beam. Similarly, the face of the
speaker is positioned with respect to the speaker hole by
aligning the forehead and chin to the box. The physiology of
each speaker is different, resulting in variations of the posi-
tion of the mouth relative to the light sheet. Hence, the

o
n
A

u
g
u
st 2

7
, 2

0
2
0

h
ttp

://a
d
va

n
ce

s.scie
n
ce

m
a
g
.o

rg
/

D
o
w

n
lo

a
d
e
d
fro

m

http://www.advances.sciencemag.org/

http://advances.sciencemag.org/

Amy
Highlight

Amy
Highlight

Amy
Highlight

Amy
Highlight

Amy
Highlight

Amy
Highlight

Amy
Highlight

Amy
Highlight

Amy
Highlight

First release: 7 August 2020 www.advances.sciencemag.org (Page numbers not final at time of first release) 3

droplet count reflects only a portion of all droplets, but as we
perform the experiment with same initial conditions for all
masks, the relative performance of the masks can be com-
pared. A speaker hole that is sealed around the face would
prevent the undetected escape of particles and ease compar-
ison between different speakers.

Second, the use of a cell phone camera poses certain lim-
itations on detection sensitivity, i.e., the smallest recogniza-
ble droplet size. To estimate the sensitivity, we consider the
light that is scattered by droplets passing through the laser
beam. The amount of light scattered into the camera direc-
tion depends on the wavelength of light, the refractive index
of the droplet, and its size (and shape). To estimate the light
scattering of droplets into the camera as a function of their
diameter we used the Python package PyMieScatt (12), which
is an implementation of Lorenz-Mie theory (see (13) for a re-
view). The result is visualized in Fig. 4. Panel (A) shows an
example of the scattering distribution for 532 nm light scat-
tered from a droplet of 5 μm diameter and a refractive index
of water (n=1.33). In this example, the particle size is substan-
tially bigger than the wavelength of the light (the so-called
Mie regime). Almost all the light is scattered into the forward
direction (0°) and very little into the direction of the camera
(indicated by the shaded green cone around 90°). For the
given camera acceptance angle, we display in Fig. 4 (B) the
estimated number of photons per frame scattered into the
cell phone camera aperture as a function of particle diameter.
By illuminating the camera directly with an attenuated laser
beam of known power, we determine the detection sensitiv-
ity. A minimum of about 75 photons (on a single camera
pixel) or about 960 photons (spread over several pixels) per
frame were required for the camera to detect a droplet (for
details on the detection characterization, see the supplemen-
tary materials). Both detection thresholds are indicated by
horizontal black lines in and the red shaded area in Fig. 4
(B). The more conservative detection threshold corresponds
to a minimum detectable droplet size of 0.5 μm. The main
limitation is the low collection efficiency of our small camera
aperture – we currently capture only 0.01% of the full solid
angle. An increased collection efficiency is possible with a
larger relay lens in front of the camera, but this would come
at the cost of a reduced field of view.

Third, the use of a single cell phone camera also limits the
achievable size resolution (currently 120 μm/pixel), given the
large field of view that is required to image as many droplets
as possible. This makes it unfeasible to directly measure the
size of small (aerosol) droplets in our setup. However, while
we cannot measure the size of droplets at or below the pixel
resolution, we can still detect and count the smaller droplets,
down to the sensitivity limit described above. For very large
particles, the limited dynamic range of the camera also poses
a challenge for determining the size, since pixels easily

saturate and hence distort the shape of the recorded droplet.
We want to point out that neither the limited pixel resolution
nor the saturation affect the particle counts presented in Fig.
3. Choosing a higher quality camera and a smaller field of
view, combined with a funnel setup to guide droplets toward
the imaging area, would reduce the minimum observable
size; so would approaches which use camera arrays to im-
prove resolution without sacrificing sensitivity or field of
view (14). Keeping in mind these sizing limitations, we can
still estimate the size distribution for the larger droplets (see
supplementary figure S5 for a qualitative size plot), which
presents some interesting observations such as the fleece per-
formance mentioned earlier.

We should point out that our experiments differ in several
ways from the traditional methods for mask validation, such
as filtration efficiency of latex particles. As is apparent from
the neck fleece study, liquid filtration (and subsequent parti-
cle size reduction) are more relevant than solid filtration. In
addition, our method could inform attempts to improve
training on proper mask use and help validate approaches to
make existing masks reusable.

In summary, our measurements provide a quick and cost-
effective way to estimate the efficacy of masks for retaining
droplets emitted during speech for droplet sizes larger than
0.5 μm. Our proof-of-principle experiments only involved a
small number of speakers, but our setup can serve as a base
for future studies with a larger cohort of speakers and checks
of mask performance under a variety of conditions that affect
the droplet emission rate, like different speakers, volume of
speech (15), speech patterns (16), and other effects. This
method can also test masks under other conditions, like
coughing or sneezing. Improvements to the setup can in-
crease sensitivity, yet testing efficiency during regular breath-
ing likely will require complementing measurements with a
conventional particle sizer. A further area of interest is the
comparison of mask performance between solid particles and
droplets, motivated by the observed liquid droplet breakup in
the neck fleece and mask saturation by droplets, necessitat-
ing exchange in regular clinical practice.

Materials and Methods
The optical setup we employed was recently used to demon-
strate expulsion of liquid droplets during speech and for
characterization of droplet residence times in air (1, 2). A
schematic of the setup is shown in Fig. 1. In short, a light
sheet was shined through an enclosure where light scattering
from particles traversing the light sheet was detected with
the camera. To form the light sheet, a cylindrical lens trans-
formed a green laser beam into an elliptical profile, which
was directed through the enclosure. The laser source was a
scientific pump laser (Millennia, Spectra-Physics; power 2 W,
wavelength 532 nm), but suitable green lasers of similar

o
n
A

u
g
u
st 2

7
, 2

0
2
0

h
ttp

://a
d
va

n
ce

s.scie
n
ce

m
a
g
.o

rg
/

D
o
w

n
lo

a
d
e
d
fro

m

http://www.advances.sciencemag.org/

http://advances.sciencemag.org/

Amy
Highlight

First release: 7 August 2020 www.advances.sciencemag.org (Page numbers not final at time of first release) 4

powers are available for less than US $100; the scientific la-
sers have better specifications (higher beam pointing and in-
tensity stability, better beam profile), but these advantages
are irrelevant in this application. The light sheet at the center
of the enclosure had a thickness of 4.4 mm and a vertical size
of 78 mm (Gaussian 1/e2 intensity beam widths). The enclo-
sure (L x W x H: 30 cm × 30 cm × 35 cm) was constructed out
of (or lined with) black material to minimize stray light. The
sides of the box had slits for entry and exit of the light sheet.
The front of the box had an 18 cm diameter hole for the
speaker – large enough for a person wearing a mask to speak
into the box but small enough to prevent the face (or mask)
from reaching the light sheet. In to clear droplets from
the box between experiments, laminar HEPA-filtered air was
continuously fed into the box from above through a duct of
cross section 25 cm × 25 cm. The supplied air was being ex-
pelled through the light sheet slits and the speaker hole. A
slight positive pressure in the box cleared droplets and pre-
vented dust from entering into the box from outside. On the
back of the box, a cell phone (Samsung Galaxy S9) was
mounted at a distance of 20 cm from the light sheet. Using
the Android app “Open Camera” the frame size was set to
1920 × 1080 pixels, the focal distance to 20 cm, the exposure
time to 1/50 s, and the frame rate to 30/s. At this focal dis-
tance, each camera pixel recorded an area of 120 μm × 120
μm at the position of the light sheet.

For each trial, the camera recorded scattered light from
particles in the laser beam before the speech (~10 s), during
speech (~10 s), and for a period of droplet clearing (~20 s).
The speech consisted of five repetitions of the phrase “Stay
healthy, people,” spoken by a male test person with a strong
voice but without shouting. Each trial was repeated ten times
and the speaker drank a sip of water in between to avoid de-
hydration. Furthermore, for the masks that showed substan-
tial amounts of detected particles (knitted, cotton, fleece, and
bandana), we conducted additional tests by repeatedly puff-
ing air from a bulb through the masks, rather than speech
from an experimenter. These control trials with air puffs con-
firmed that we recorded droplets emitted by the speaker, not
dust from the masks.

The goal of the analysis is to compare the efficacy of dif-
ferent masks by estimating the total transmitted droplet
count. Toward this end, we need to identify droplets in the
video and discriminate between droplets and background or
noise. For convenience, analysis of the videos was performed
with “Mathematica” (Wolfram Research) but use of a com-
mercial package does not pose any general restriction since
almost every high-level programming language (e.g., Python)
offers the same functionality. From all videos, we removed a
weak background that originated from the light sheet itself
and from stray light and diffuse reflections from the experi-
menter’s face. We then binarized all frames with a common

threshold that discriminates between scattered light from
droplets and background signal and/or noise. Then, a feature
detection algorithm is applied to each frame, which returns
the center of mass positions, and major axis and minor axis
length of the best-fit ellipse for every droplet. Note that the
major and minor axis returned by the algorithm are not a
direct measure of the droplet size, but a measurement of the
amount of light scattered by the particle into the camera ap-
erture (binary diameter). Furthermore, the major axis length
is increased due to particle motion during the camera expo-
sure time. Due to the small dynamic range of the camera (8-
bit), most droplets saturate the camera. However, the axes
lengths returned by the algorithm can still be used for a qual-
itative droplet size estimation: a bigger droplet scatters more
light than a smaller droplet. This insight is important to in-
terpret the result of the neck fleece. The neck fleece has a
larger transmission (110%, see Fig. 3 (A)) than the control
trial. We attribute this increase to the neck fleece dispersing
larger droplets into several smaller droplets, therefore in-
creasing the droplet count. The histogram of the binary di-
ameter for the neck fleece supports this theory (see Suppl.
Fig. S5).

If a droplet passes through the light sheet in a time
shorter than the inverse frame rate, it will appear only in a
single video frame. However, if the droplet spends more time
in the light sheet, the droplet will appear in multiple frames.
To avoid double-counting droplets in consecutive frames, we
use a basic algorithm to distinguish between single-frame
particles and multi-frame trajectories. The algorithm com-
pares the distance between droplets in consecutive frames
and assigns two droplets to a trajectory if their distance is
smaller than a threshold value or counts them as individual
droplets if their distance is larger than the threshold. The
threshold value was empirically chosen to be 40 pixels. An
example result of the algorithm is shown in Supplementary
Fig. S6, which shows a projection of 10 consecutive frames.
Every droplet recognized by the algorithm is highlighted by
an ellipsoid, labeled with the frame number. Droplets that
belong to the same trajectory are highlighted in the same
color.

REFERENCES AND NOTES
1. V. Stadnytskyi, C. E. Bax, A. Bax, P. Anfinrud, The airborne lifetime of small speech

droplets and their potential importance in SARS-CoV-2 transmission. Proc. Natl.
Acad. Sci. U.S.A. 117, 11875–11877 (2020). doi:10.1073/pnas.2006874117
Medline

2. P. Anfinrud, V. Stadnytskyi, C. E. Bax, A. Bax, Visualizing Speech-Generated Oral
Fluid Droplets with Laser Light Scattering. N. Engl. J. Med. 382, 2061–2063
(2020). doi:10.1056/NEJMc2007800 Medline

3. N. L. Belkin, The evolution of the surgical mask: Filtering efficiency versus
effectiveness. Infect. Control Hosp. Epidemiol. 18, 49–57 (1997).
doi:10.2307/30141964 Medline

4. A. S. Dharmadhikari, M. Mphahlele, A. Stoltz, K. Venter, R. Mathebula, T. Masotla,
W. Lubbe, M. Pagano, M. First, P. A. Jensen, M. van der Walt, E. A. Nardell, Surgical
face masks worn by patients with multidrug-resistant tuberculosis: Impact on
infectivity of air on a hospital ward. Am. J. Respir. Crit. Care Med. 185, 1104–1109

o
n
A

u
g
u
st 2

7
, 2

0
2
0

h
ttp

://a
d
va

n
ce

s.scie
n
ce

m
a
g
.o

rg
/

D
o
w

n
lo

a
d
e
d
fro

m

http://www.advances.sciencemag.org/

http://dx.doi.org/10.1073/pnas.2006874117

http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=PubMed&list_uids=32404416&dopt=Abstract

http://dx.doi.org/10.1056/NEJMc2007800

http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=PubMed&list_uids=32294341&dopt=Abstract

http://dx.doi.org/10.2307/30141964

http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=PubMed&list_uids=9013247&dopt=Abstract

http://advances.sciencemag.org/

First release: 7 August 2020 www.advances.sciencemag.org (Page numbers not final at time of first release) 5

(2012). doi:10.1164/rccm.201107-1190OC Medline
5. “Advice on the use of masks in the context of COVID-19” (WHO Reference Number:

WHO/2019-nCov/IPC_Masks/2020.4, World Health Organization, 2020).
6. M. Klompas, C. A. Morris, J. Sinclair, M. Pearson, E. S. Shenoy, Universal Masking in

Hospitals in the Covid-19 Era. N. Engl. J. Med. 382, e63 (2020).
doi:10.1056/NEJMp2006372 Medline

7. D. K. Chu, E. A. Akl, S. Duda, K. Solo, S. Yaacoub, H. J. Schünemann, COVID-19
Systematic Urgent Review Group Effort (SURGE) study authors, Physical
distancing, face masks, and eye protection to prevent person-to-person
transmission of SARS-CoV-2 and COVID-19: A systematic review and meta-
analysis. Lancet 395, 1973–1987 (2020). doi:10.1016/S0140-6736(20)31142-9
Medline

8. N. H. L. Leung, D. K. W. Chu, E. Y. C. Shiu, K.-H. Chan, J. J. McDevitt, B. J. P. Hau, H.-
L. Yen, Y. Li, D. K. M. Ip, J. S. M. Peiris, W.-H. Seto, G. M. Leung, D. K. Milton, B. J.
Cowling, Respiratory virus shedding in exhaled breath and efficacy of face masks.
Nat. Med. 26, 676–680 (2020). doi:10.1038/s41591-020-0843-2 Medline

9. D. Bunyan, L. Ritchie, D. Jenkins, J. E. Coia, Respiratory and facial protection: A
critical review of recent literature. J. Hosp. Infect. 85, 165–169 (2013).
doi:10.1016/j.jhin.2013.07.011 Medline

10. J. J. Y. Ong, C. Bharatendu, Y. Goh, J. Z. Y. Tang, K. W. X. Sooi, Y. L. Tan, B. Y. Q.
Tan, H.-L. Teoh, S. T. Ong, D. M. Allen, V. K. Sharma, Headaches Associated With
Personal Protective Equipment – A Cross-Sectional Study Among Frontline
Healthcare Workers During COVID-19. Headache 60, 864–877 (2020).
doi:10.1111/head.13811 Medline

11. A. Konda, A. Prakash, G. A. Moss, M. Schmoldt, G. D. Grant, S. Guha, Aerosol
Filtration Efficiency of Common Fabrics Used in Respiratory Cloth Masks. ACS
Nano 14, 6339–6347 (2020). doi:10.1021/acsnano.0c03252 Medline

12. B. J. Sumlin, W. R. Heinson, R. K. Chakrabarty, Retrieving the aerosol complex
refractive index using PyMieScatt: A Mie computational package with visualization
capabilities. J. Quant. Spectrosc. Radiat. Transf. 205, 127–134 (2018).
doi:10.1016/j.jqsrt.2017.10.012

13. G. Gouesbet, G. Gréhan, Generalized Lorenz-Mie Theories (Springer International
Publishing, ed. 2nd, 2017).

14. L. Zhong, P. Konda, M. Harfouche, R. Horstmeyer, in Imaging and Applied Optics.
(Optical Society of America, 2019).

15. S. Asadi, A. S. Wexler, C. D. Cappa, S. Barreda, N. M. Bouvier, W. D. Ristenpart,
Aerosol emission and superemission during human speech increase with voice
loudness. Sci. Rep. 9, 2348 (2019). doi:10.1038/s41598-019-38808-z Medline

16. S. Asadi, A. S. Wexler, C. D. Cappa, S. Barreda, N. M. Bouvier, W. D. Ristenpart,
Effect of voicing and articulation manner on aerosol particle emission during
human speech. PLOS ONE 15, e0227699 (2020).
doi:10.1371/journal.pone.0227699 Medline

ACKNOWLEDGMENTS

We thank Mathias Fischer for providing the sketch in Fig. 1, and Shannon Eriksson
and Jake Lindale for valuable discussions. Funding: This project has been made
possible in part by grant number 2019-198099 from the Chan Zuckerberg
Initiative DAF, an advised fund of Silicon Valley Community Foundation, and by
internal funding from Duke University through the Advanced Light Imaging and
Spectroscopy (ALIS) facility. Author contributions: M.C.F. and E.P.F. performed
the experiments, D.G. performed the data analysis, I.H. and E.W. procured the
masks, and W.S.W. provided expertise. M.C.F supervised the project. All authors
were involved in data interpretation and manuscript preparation. Competing
interests: A US provisional patent application has been filed by Duke University
on …