To be fair, if someone made a voice encoding only with that had a lot of error correction bits, it would probably work at much longer distances. Some of the codecs are 2 kilobits per second for human voice? That's got to have a way better margin for the same channel bandwidth than analog decoded by the human brain. This way we get the digital advantage and lossy compression.
MELP targets 2.4kbps, and there are later examples for even lower bitrates (e.g. 1.2kbps, 600bps)
One thing to keep in mind is that a lot of these codecs or technologies are designed to operate at very low bit rates because the signal carrier is required to be capable of operation in "contested environments" where jamming and/or other environmental effects are present (and throughput potential is the tradeoff for assurance).
Especially for speech, there are aggressive and impressive algorithms that can turn a trickle of bits into understandable voice.
> Digital radio remains working at much worse SNR than intelligible fuzzy analog.
Your opus codec needs to demodulate the signal first.
So long as you can get pops of decodable signal, you'll be able to understand what's being said. Using a more robust modulation scheme would also go a long way in improving reach.
AM gets it's range benefits not from the modulation used, but from the frequencies it occupies.