Because they talk over noisy, distorted channels that have about 3 kHz bandwidth, in environments that are noisy to begin with?
You can talk to your PC in 16 bit (or more) audio sampled at 48 kHz, with a decent microphone, in a relatively quiet environment. Moreover, it could be tailored to understand your voice specifically, not just anyone saying "U".
When you're combining letter names with other commands, even in perfect recording conditions, there will be confusion. The command "back Q" sounds very very close to "back U", but "back quench" is different than "back urge".
In either case, you're using the same number of syllables, so there isn't any speed lost. You just have to spend a day learning the alphabet.
"back U", or perhaps more pertinently "black U", is not very close to "black Q". They are phonetically distinct. "black U" has no stop; in "black Q" there is a stop between the words. The final plosive of "black" is deleted so it becomes something similar to "bla ' kyu", whereas the other one is "blakyu".
In 2023, if whatever you're using doesn't listen for stops, here is a dime, get yourself a better speech recognizer?
I'm not learning any phonetic alphabet other than the ICAO standard one.
If I ever have to use that one, I will look like a fool if that nonstandard one starts slipping in by mistake.
Because they talk over noisy, distorted channels that have about 3 kHz bandwidth, in environments that are noisy to begin with?
You can talk to your PC in 16 bit (or more) audio sampled at 48 kHz, with a decent microphone, in a relatively quiet environment. Moreover, it could be tailored to understand your voice specifically, not just anyone saying "U".