# How is voice sound on TV or phone distorted through a face mask?

After a year of listening to people talk through face masks, it appeared to me that the sound is not merely muffled, but it has a very specific and recognizable tinge and texture to it.

It reminds me less of sound through a muffling pillow, and more of old speech recordings from the 1950s.

This is apparent whether you listen, today, to recorded speech on TV or radio or through the phone. Sometimes you hear it in person too.

I suspect that the sound is not just muffled or low-pass filtered, but that there is more happening. Perhaps it's a band-pass? A notch? Are there non-linear or time-variant effects from turbulence due to the interaction between plosives and the fabric?

Has anyone looked into this?

• Hmm. I think you could have extra muffling of the plosives just from spatially sensitive linear effects -- I could be wrong, though. May 27 at 17:08
• One thing you should not forget is that our perception can fool us in so many ways. In particular, listening for a long period of time to a certain sound will make our brains perceive strange tonalities not noticed before. That doesn't mean they are there, in the same manner that staring at the walls of a cave for a long time can make people think they spot animals in the cracks and reliefs. They may paint over them what they think they saw, but that was simply their mind playing tricks. Not lastly, a mask is not LTI because it's a very dynamic element, moving with the breath and wind. May 27 at 17:50
• @aconcernedcitizen while fully agreeing with what you write (perception and LTI), are you also suggesting that the timbre of a masked voice that I believe to hear is a result of such a psycho-acoustic effect, or would you lean more towards a physical explanation? May 28 at 17:50
• @TimWescott yes that "extra" or at least different filtering provides for a time varying filter driven/modulated by air flow and the impedance at the oral cavity (See discussion under ZR Han's answer below). That's why I also think it's more than just a (low-pass-ish) linear absorption model. May 28 at 17:53
• @P2000 Here's a thought: say both you and me are listening to a speech through a mask, and both of us are saying that the sound appears to have a certain "colour". What would your thought be: that I am hearing the exact same colour as you, or that it differs according to each one's perception? There's no denying that the masks do alter the sound, but how we hear it is based on our perception. It's just like the old saying: "de gustibus et coloribus non est disputandum". May 28 at 19:21

There are many researches on the acoustic effect of masks since the pandemic. You can find them on the website of The Journal of the Acoustical Society of America and most of them are free to access.

Masks have selectivity on both frequency and spatial.

In frequency domain, masks react as low-pass filters that attenuate high frequencies significantly. Interestingly, different research gives different results. They are saying 1 kHz$$^{[1]}$$, 2 kHz$$^{[2]}$$ and 3 kHz$$^{[3]}$$ maybe due to different understanding about "significantly".

In spatial domain, masks lead to reduced frontal sound radiation compared to the lateral or rear radiation at high frequencies$$^{[2]}$$.

Researchers also tested the effect of microphone placement. They placed microphones on the lapel, cheek, forehead, and next to the mouth$$^{[1]}$$. These differences are mostly caused by the spatial radiation pattern with a mask.

And of course, different masks have different acoustic effect according to their materials, weaves and types.

References

• Thank you for providing ample references. They all confirm that there is attenuation, some low pass effect (with exceptions) and modify the radiation pattern. What I feel is missing in the research is whether the speaker plus mask and a filtered voice signal both result in the same human perception as a real mouth with mask (esp. plosives and open mouth vowels). It would be insightful to filter a voice recording with a few of the curves provided to hear if it yields a similar timbre as a masked mouth, and I wouldn't be surprised if they don't match. Thoughts? May 28 at 4:57
• @P2000 If a loudspeaker sounds like a real mouth without mask, then putting a mask on it should make it sound like a real mouth with mask. If a voice signal is recorded near the eardrum when a masked person speaks to you, and then reproduced by an ideal earphone, they should be the same. That's similar to the fundamental of head-related transfer function (HRTF). May 28 at 5:37
• I think because the mask is in the acoustic near field, that assumption does not hold. There is an interaction that is not modelled by cascaded transfer functions. The mask changes the impedance at the opening, the air density (humidity) in the mouth, and generates noise from airflow, all which a cascaded independent mask filter cannot model. I think this is overseen or dismissed, but it does matter in the sound distortion. Yet, I have no measurements or references for this theory...! Would you agree these are plausible contributors to effects omitted from the model? May 28 at 14:27
• Yes there are some factors that cannot be modeled by an LTI system such as nonlinearity and noise. But I think the impedance changing and the humidity can be modeled to some extent. Acoustical impedance is linked to electronic impedance and is analogous to the RLC circuit. Absorption of the sound wave due to air humidity is often modeled as a low pass filter. Of course the real physical process is very complicated and no mathematical model can conclude the whole process, but we can use a model which is close enough. May 28 at 14:48
• Agreed, and I think this goes towards the point that a mask in front of a loudspeaker is not the same as a masked mouth, and the subtle differences are indeed audible (timbre, texture...). Thank you for this discussion. May 28 at 17:47