Not sure if this has been discussed, but the Baofeng DM32UV can do 256 bit DMR encryption at a seriously impressive price point and with much improved signal quality/supriousness/etc over their previous slop.
Would be very useful post-habbening methinks...
Since this thread contains "real HAMs" and by extension the "FCC police" people, I'm sorry FCC. I won't do it again.
Many Tytera DMR products like the TYT MD-UV390 and TYT MD-UV990 also support AES-256 encryption
That said, there is no way to audit the implementation of such in their firmware without extracting the blob and going through it yourself. You are betting your luck on a Chinese manufacturers competency. While I do like the Tytera brand and they have great quality for their handhelds compared to most Chinese manufacturers I still wouldn't trust something like that where it may be critical.
The best "SHTF" comms setup for
voice is any analog radio, set to the lowest bandwidth, with the lowest amount of power that can facilitate communications between you and the people you need to talk to. Also with the highest amount of directionality (Yagi's are a good investment)
The radio is probably the least important part of the equation. Knowing what to say and how to say in the shortest amount of time is far better and works with any piece of equipment.
A future project I plan to work on is a local
whisper.cpp speech to text model running on a Radxa Zero SBC that transcribes audio when a PTT mic is activated into text, and then sends it as data rather than voice. The incoming messages would then be played back using text to speech and also optionally shown on the E-ink display (they are great as they use negligible power, and emit no light).
It would be able to work over any medium, and DMR radios that already have data burst / SMS capability with KISS support would work well. But I'm also eyeing 2.4 GHz LoRa with S2X180 boards.
A voice message could be compacted down to a less than a second on air with routing capabilities, and it could even be sent under the noise floor.
The most difficult part of the project would be the form factor and power consumption. whisper.cpp's smallest model uses a full gig of memory.
Transcrption errors could be mitigated by using biases towards a predefined dictionary, like the phonetic alphabet and decoder tuning t,hat whisper.cpp supports out of the box.