Welcome,
1) Dynamically generate (relatively) small grammar sets (or language models as they’re called here I guess)
Language models (which are statistical models) and grammars (which are rulesets) are actually two different things, but both are supported by OpenEars. You can read more about both in the documentation for PocketsphinxController and LanguageModelGenerator if you’re interested. Choosing between them is generally one of the most important UX design decisions for a speech UI.
2) After a recognition event, the previous grammar is obsolete and I need to be able to quickly trash the current models and insert new ones
OK, this is the standard use case that the language model and grammar generation API was designed for.
Basically, I’m having trouble when I try to load new language models and it seems to take longer than necessary to load models in the first place.
This doesn’t really take an easily-perceptible amount of time, so if you are ever finding yourself waiting for a language model or grammar to be generated, that is a suggestion that there is some other issue. What is your indication that it is language model generation that is causing a delay versus something else?
It doesn’t sound like you need a plugin necessarily since your use case is the standard one. The only thing that sounds unusual is being concerned with the models or grammar writing out (what is the concern there?), and the issue with the delay or perception of delay due to some other cause. We can troubleshoot the delay and see what it is that is causing that symptom.