Yeah, true. They've made some pretty impressive perceptual voice codecs that take very little data, and using that as the carrier for transmitting data would eat into that quite a bit more. Though, I find steganographic methods quite fascinating and often catch myself pondering ways to exchange data privately where efficiency is a small aspect of the overall goal of the communication channel. I suppose "thought experiment" is a good description.

The most effective/efficient way to do E2EE voice is definitely to use an app that supports voice and transfers via the data channel.