Hi,
There are a lot of discussions on these forums and Stack Overflow as well about the basic difficulty in offline speech recognition for recognizing contextless and rhyming one-syllable utterances like letters and numbers (there is actually a forum topic about this on the front page of the OpenEars forum topic list right now, or you can search for the keyword ‘numbers’). So, I can’t help you with your basic premise of an app which has required functionality based on this not-particularly-solved problem in speech recognition.
This is very frustrating and I’ve put tons of time into implementing this.
First apps are time-consuming and frustrating, and it’s normal to attribute that to whatever non-trivial core technology you’re using for it. I have also put serious time into supporting you although most of your questions are about your programming learning process, and that’s OK (we’ve all been there), but misattributed venting of your development stress to me isn’t a constructive path to go down (although we’ve all been there).
There are a couple of issues with your language model that could be improved. Your goal of only having utterances that take the form of:
“THE NUMBER IS _______”
Is not something that can be expressed by a statistical language model, which deals with probabilities rather than requirements. You want a grammar, which is a way of expressing an absolute rule for a matching utterance. Correct use of a grammar will limit the search field a bit more for your number at the end and could perhaps give a small improvement. This is a starting point for reading about grammars:
https://www.politepix.com/2014/04/10/openears-1-7-introducing-dynamic-grammar-generation/
For difficulties which arise when implementing this as a grammar, don’t forget to search these forums first for other developer’s similar questions that may have been answered. Good luck!