Currently the handling of spoken numbers in Speech to text is making it very annoying for my use-case.

I am using LUIS to identify the following entities:

{Group} {Flight} {Plane}

Where Flight and Plane are individual single digit numbers. For example (Using the text representation of numbers when spoken to avoid confusion)

"Enfield one two" spoken gets rendered to text and passed into LUIS as "Enfield 12" which round-tripping back to the text-to-speech is synthesized as "Enfield twelve". the lack of space between each digit also makes it impossible to assign them to individual entities (even using lists and regex types).

The only semi-reliable way I have of getting 1 2 (Note the space) is by leaving an unnaturally long gap between the two numbers and even this is usually put in text as "one, to"

I have decided that the only thing to do is change the LUIS template to {Group} {Flight} which reliably gets me "Enfield 12" then split the numbers that make up the {Flight} entity to get the individual components. Not ideal but seems to work and I just have to insert a space when going back via TTS.