Aerosol emission and superemission during human speech increase with voice loudness

Abstract

Mechanistic hypotheses about airborne infectious disease transmission have traditionally emphasized the role of coughing and sneezing, which are dramatic expiratory events that yield both easily visible droplets and large quantities of particles too small to see by eye. Nonetheless, it has long been known that normal speech also yields large quantities of particles that are too small to see by eye, but are large enough to carry a variety of communicable respiratory pathogens. Here we show that the rate of particle emission during normal human speech is positively correlated with the loudness (amplitude) of vocalization, ranging from approximately 1 to 50 particles per second (0.06 to 3 particles per cm³) for low to high amplitudes, regardless of the language spoken (English, Spanish, Mandarin, or Arabic). Furthermore, a small fraction of individuals behaves as "speech superemitters," consistently releasing an order of magnitude more particles than their peers. Our data demonstrate that the phenomenon of speech superemission cannot be fully explained either by the phonic structures or the amplitude of the speech. These results suggest that other unknown physiological factors, varying dramatically among individuals, could affect the probability of respiratory infectious disease transmission, and also help explain the existence of superspreaders who are disproportionately responsible for outbreaks of airborne infectious disease.

Figure 1

Representative raw data in which a participant (F4) said /ɑ/ for 5 seconds, followed by 15 seconds of nose breathing, repeated 6 times at approximately the same loudness. (A) The amplitude (arb. units) recorded by the microphone versus time. Magnification shows 13 ms of the waveform with fundamental frequency of F₀. (B) The corresponding number/concentration of particles measured by the APS versus time.

Figure 2

Particle emission rate/concentration while saying /ɑ/ at 8 different amplitudes, repeated 6 times at each amplitude. (A) Particle emission rate/concentration versus root mean square amplitude, A_rms (arb. units) for a representative participant (F4). Solid line is the best linear fit, with correlation coefficient ρ = 0.932 and Pearson’s p value = 5.9 × 10⁻²². (B) Corresponding particle size distribution for the data presented in (A). (C) Aggregated particle emission rate/concentration versus root mean square amplitude, A_rms (arb. units) for 10 participants, 5 males (denoted as M1 to M5) and 5 females (denoted as F1 to F5). There are 8 data points for each participant, each representing the average of repeating /ɑ/ six times at approximately the same voice amplitude (cf. Fig. 1). Solid line is a power law fit with exponent 1.004, correlation coefficient ρ = 0.774 and Pearson’s p value = 3.8 × 10⁻¹⁷.

Figure 3

Particle emission rate/concentration while reading a passage of text aloud (the “Rainbow” passage), at three different loudness levels. (A) Superimposed representative recordings of amplitude (arb. units) for an individual (F4) reading the passage at three different voice amplitudes, and (B) the corresponding number/concentration of particles measured by the APS versus time. Color code same as in (A). (C) Particle emission rate/concentration as a function of root mean square amplitude, A_rms, for 10 participants. There are 3 points for each person, representing 3 voice amplitudes, color code same as Fig. 2C. Solid line is a power law fit with exponent 0.96, correlation coefficient ρ = 0.865 and Pearson’s p value = 6.8 × 10⁻¹⁰. (D) Representative particle size distribution for the one individual (F4).

Figure 4

Histogram of particle emission rate/concentration at voice amplitude of 0.1 (approximately 85 dB). (A) For saying /ɑ/, with median of M = 4.3 particles/s, mean of m = 4.8 particles/s and standard deviation of σ = 3.0 particles/s. (B) For reading an English passage (10 people read the “Rainbow” passage and 30 people read chapter 24 of “The Little Prince”) with median of M = 2.5 particles/s, mean of m = 3.4 particles/s and standard deviation of σ = 2.7 particles/s. Particle emission rates larger than m + σ are labeled superemitters. Red curves are lognormal fits found via nonlinear regression.

Figure 5

Comparison of (A) emission rate/concentration and (B) corresponding geometric mean diameters of particles emitted during various modes of breathing versus speech at different loudness levels. “Nose” denotes normal nasal breathing; “Mouth” denotes normal mouth breathing; “Deep-Fast” denotes deep, slow nasal inhalation followed by fast mouth exhalation; “Fast-Deep” denotes fast nasal inhalation followed by deep (i.e., slow and prolonged) mouth exhalation. “Quiet”, “Intermediate”, and “Loud” denote loudness levels while reading aloud a passage of text (“Rainbow” passage) at respective amplitudes. Red lines indicate medians, while bottom and top of blue boxes indicate the 25^th and 75^th percentiles respectively; sample size is n = 10. Outliers (defined as values that exceed 2.7 standard deviations) are indicated with red plus signs. Note that the 2 outliers for speech in (A) are a different individual (F4) than the two outliers observed for nose and fast-deep breathing (M24 and M5 respectively). Scheffe groups are indicated with letters; groups with no common letter are considered significantly different with p < 0.05, cf. Supplementary Table S1. Note that (A) has different scales above and below the break.