Yeah. You can't just produce random white sound[1] to cover the signal either, because speech has enough redundancy to still communicate over a static-y soundscape. The term here is that white noise lacks "diffusion"--analogous to the ECB vulnerability (see the picture of the encrypted Linux penguin image on this page[2]). I can't think of a good way to provide diffusion for sound--a better cover might be to provide alternate signals, i.e. a few layers of random human sounds spoken in a deepfake-d version of the voice of the person speaking.

I'm skeptical even this wouldn't have vulnerabilities however, as history has shown us that cryptography which has not been formally proven is often broken.

[1] I'm using the word "sound" here to differentiate the noise in the auditory sense from noise in the "signal versus noise" sense.

[2] https://en.wikipedia.org/wiki/Block_cipher_mode_of_operation...