Abstract
To prevent the transmission of Covid-19 the US Centers for Disease Control recommends using a face mask in public. Converging evidence examining the acoustics of masked speech shows that different styles of masks have differential effects on sound attenuation. The purpose of this study was to compare speech intelligibility in different mask conditions (no mask, disposable surgical mask, cloth mask, and N95 respirator). Four native English speakers recorded unpredictable sentences, which avoid the confound of contextual predictability. Sentences were mixed with multitalker babble to simulate the noise experienced by listeners during activities of daily living. Forty-one listeners heard 24 pseudorandomized unpredictable sentences from the four mask conditions and typed what they heard. We measured intelligibility as the percentage of whole words correctly perceived. Acoustic analysis revealed that all masks filter the signal, with the greatest overall effect for the N95 respirator. A mixed model two-way ANOVA showed a significant effect of mask condition, driven by the low percent words correct for N95 respirators. Our error analysis revealed that listeners more often provided no response for the N95 respirator, but supplied phonetic approximations for surgical and cloth masks.