Face Mask Effects on Acoustic Features and Intelligibility of Mandarin Chinese Speech.

Purpose: The goal of this study was to investigate how face masks influenced the acoustic features of Chinese running speech in both temporal and spectral domains and how the intelligibility of the speech with face masks was affected in quiet and multitalker babbles. The relationship between the acoustic features and speech intelligibility was also examined.

Method: In Experiment 1, Mandarin Chinese sentences were recorded by 24 native Mandarin Chinese speakers wearing a surgical mask, a KN95 mask, or not wearing a mask and temporal modulation (TM) depth; speaking rate, spectral tilt, and average value and standard deviation of fundamental frequency (F0) were then examined. In Experiment 2, the intelligibility of these recorded sentences were assessed in quiet and multitalker babble with the signal-to-noise ratios of -2 and -5 dB. To further examine the possible causal relationship between the impacted acoustic variables and speech intelligibility under different mask wearing conditions, the acoustic and speech intelligibility data were analyzed in a stepwise regression.

Results: Results showed that both the KN95 and surgical masks produced significantly smaller TM depth compared to the no-mask condition. In terms of speaking rate, participants spoke faster with face masks than without a mask, whereas there was no significant difference between the KN95 and surgical mask. Additionally, spectral tilt was significantly shallower for the two face masks compared to the no-mask condition. Regarding F0, the mean F0 was higher with the KN95 mask than the surgical mask and no mask, while the standard deviation of F0 was lower in the two mask conditions than the no-mask condition, with no significant difference between the two types of masks. In addition to these acoustic differences, speech intelligibility in noise was significantly lower for the two mask conditions than the no-mask condition, with no significant difference between the KN95 and surgical masks, whereas there was no significant effect of face masks on speech intelligibly in quiet. Finally, the relationship between acoustic features and speech intelligibility showed that, under noise conditions, TM depth, spectral tilt, and F0 dynamics (e.g., standard deviation) were significantly correlated with speech intelligibility, while speaking rate and mean F0 were not.

Conclusions: Acoustically, face masks led to smaller TM depth, slower speaking rate, shallower spectral tilt, higher mean F0 and smaller standard deviation of F0 in Mandarin Chinese running speech, and perceptually resulted in lower speech intelligibility in noise, but had no impact on speech intelligibility in quiet. Findings also suggest that certain acoustic characteristics (e.g., TM depth and spectral tilt) play important roles on speech intelligibility, especially in challenging listening conditions.