Effects of surgical masks on aerosol dispersion in professional singing

The study addresses how surgical masks affect spatial and temporal dispersion of aerosols and droplets during professional singing, a known high-risk activity for SARS-CoV-2 transmission. Aerosols (≤5 µm) can remain suspended and spread convectively, while droplets (≥5 µm) pose risk at close range. Singing produces substantially more aerosols and droplets than breathing or speaking and has been linked to superspreading events. Prior work often measured particle counts but not full 3D dispersion. The study aims to determine where the highest aerosol concentration region is during singing and to test how a surgical mask alters dispersion and droplet expulsion, to inform safer choir practices during airborne viral pandemics.

Existing literature indicates speaking emits ~3x more aerosol than breathing, while singing can emit ~3–100x more than speaking, depending on conditions. Prior outbreaks in choirs underscore risk. Mask studies show reductions in emitted particles and transmission, particularly with surgical/KN95 masks, but leakage at cheeks and nose allows aerosol escape. Droplet travel distances vary with activity and loudness, with coughing droplets detected up to 8 m; droplets follow ballistic trajectories and quickly settle. Visualization methods (smoke, laser light sheet, PIV, Schlieren) have been used mostly for breathing, coughing, sneezing, and speaking, often along the main expulsion direction only. There is a gap in 3D analysis of singing and mask effects on dispersion.

Design: Experimental study with 10 professional BR Choir singers (5 female, 5 male; age 44 ± 11 years; nonsmokers; no pulmonary symptoms). Ethical approval 20-395. Task: Sing a passage of Beethoven's 9th Symphony, "Ode of Joy," in D major (starting on F#3 for males, F#4 for females) for ~6–10 s, performed twice: without and with a standard surgical mask. Hydration encouraged. Setup 1 (Aerosol visualization): Conducted in a large studio (27 × 22 × 9 m). Singers stood on a lifting platform with measurement rods for spatial calibration. Three synchronized cameras (25 fps; two side/front with Canon lenses; one top with Fujinon wide angle) recorded the exhaled cloud. Audio captured with microphones 1.5 m from mouth. To visualize exhaled air, singers inhaled ~0.5 L of e-cigarette vapor (50% glycerin/50% propylene glycol; particle size ~250–450 nm). Inhalation volume measured via ZAN 100 spirometer coupled to the e-cigarette. After inhalation, singers moved to a marked position and performed the task, then remained stationary up to 60 s post-task. Studio conditions were controlled: black background/clothing; three flash lights for illumination; pre-task ventilation with main gate open ≥2 min and settling period ≥2 min; temperature 23.27 °C (SD 0.46), RH 46.12% (SD 0.95). Setup 2 (Droplet visualization): Within a dark chamber in the same studio, ballistic droplets (>5 µm) were illuminated by a 532 nm, 3 W laser formed into a 2 mm light sheet in sagittal orientation just in front of the mouth (laser optics via mirror arm and cylindrical lens; exit above head, sheet directed downward). Safety glasses worn; subjects not directly exposed. A Phantom v2511 high-speed camera captured at 2000 Hz within a 48 × 76 mm² ROI (1280 × 800 px; ~60 µm/px), placed ~0.7 m from the light sheet. Image Processing—Aerosols: Defined coordinate system with origin at mouth; x (front), y (left-right), z (up). Cameras 1 (side) and 3 (top) videos converted to grayscale; singer masked out. An in-house threshold-based region-growing algorithm segmented the vapor cloud per frame, yielding contours and computing maximal cloud diameters in x (dx), y (dy), and z (dz). ROIs: 260 × 270 × 180 cm (camera 1) and 190 × 270 × 180 cm (camera 3). Outliers due to non-vapor bright regions were mitigated using a moving median filter (window 30 frames) and cubic spline approximation (Matlab). Image Processing—Droplets: Particles detected per frame by grayscale threshold; tracking performed via nearest-neighbor with a 50-pixel max displacement and a deviation score (size, direction, velocity), allowing temporary disappearance at sheet boundaries. Post-processing removed likely dust based on rules: presence ≥11 frames, total displacement ≥1 pixel, mean velocity ≥2 pixels/frame. Statistics performed in Matlab and SPSS v24. Outcomes: Primary—cloud diameters dx, dy, dz over time (end of task t=0 s and t=10 s). Secondary—number of tracked droplets in ROI and number moving forward (positive x).

• Qualitative dispersion: Without a mask, the aerosol cloud propagated forward with a downward tilt, reaching up to ~1.3 m in front. With a surgical mask, aerosols were decelerated and deflected, escaping via cheek and nose leaks and remaining in the near-field around and above the head. - Quantitative cloud diameters (median across 10 singers): At t=0 s (end of task): dx: 0.85 m (without) vs 0.37 m (with); dy: 0.66 m (without) vs 1.24 m (with); dz: 1.06 m (without) vs 0.69 m (with). At t=10 s: dx: 1.11 m (without) vs 0.43 m (with); dy: 1.20 m (without) vs 1.45 m (with); dz: 1.21 m (without) vs 0.91 m (with). - The forward expansion (dx) was consistently smaller with masks, both at task end and 10 s later; transverse spread (dy) increased with masks due to lateral leakage; vertical spread (dz) was smaller in diameter with masks, but the cloud center moved higher (up to ~1.0 m above the mouth vs ~0.77 m without mask), indicating upward deflection and thermal convection. - Droplet counts in the laser-sheet ROI: Total tracked droplets decreased by 47% on average with mask vs without (Wilcoxon z = -2.497, p = 0.013, n = 10). Restricting to droplets moving forward (positive x), the reduction increased to 86% on average (Wilcoxon z = -2.803, p = 0.005, n = 10). Inter-individual variability was noted, with some singers showing higher counts under mask in total droplets, but forward-moving droplets were consistently much reduced for most singers (8/10 had <100 particles).

The findings directly address the research question by showing that surgical masks substantially reduce the forward range and momentum of aerosol emission during professional singing and markedly reduce forward-moving ballistic droplets. Masks deflect aerosol through side and nose leaks into the singer’s near-field and upward regions, decreasing exposure risk for individuals positioned directly in front. The larger transverse spread with masks reflects leakage pathways, especially affecting small aerosols that follow airflow. The upward movement of aerosols is consistent with thermal convection around the warmer human body, suggesting that ceiling-mounted ventilation can aid removal. Combined with proper ventilation, masks can allow more compact choir arrangements with reduced cross-infection risk. The results align with prior studies showing mask effectiveness and leakage dynamics, extending them to professional singing with 3D dispersion assessment. However, because aerosols remain near the singer without effective ventilation, masking alone is insufficient in closed spaces; ventilation strategies are essential to mitigate accumulation.

Surgical masks are effective in reducing the forward dispersion of aerosols during professional singing and substantially decrease ballistic droplet emission toward the front (up to 86% reduction for forward-moving droplets). Masks decelerate and deflect aerosols to the near-field around and above the singer’s head, limiting long-range exposure in front but increasing reliance on adequate ventilation to remove accumulated aerosols. While standard surgical masks are not optimized for professional singing due to articulation-induced leaks and fit issues, they remain a practical mitigation for lay choirs and congregational singing. Future work should develop singer-optimized masks minimizing leakage without compromising articulation or acoustics, and integrate ventilation assessments to quantify room-level aerosol clearance.

• Visualization used e-cigarette vapor as a surrogate, producing higher particle counts than natural exhalation; cloud boundaries depend on illumination and camera sensitivity, and regions with evaporated/diluted vapor may be undetected. - Human subject variability and unconfined flow unpredictability limit reproducibility, despite standardized protocols and 10 participants. - Droplet experiment measured relative changes only; ambient dust could not be fully excluded, and only droplets intersecting the thin laser sheet near the mouth were recorded, so totals are underestimates. - Particle sizes below ~10 µm could not be reliably sized (resolution ~60 µm/pixel), and some droplets may have escaped through mask leaks outside the ROI. - Measurements of dispersion focused on near-field dynamics; room-scale aerosol accumulation and clearance were not directly quantified.