Help with understanding language models and grammars

[aaronchantrill]

Hello all. Thank you for the fantastic work. This project is amazing!

I'm working in a fairly small domain of commands and am really trying to understand the interaction between the acoustic and language models.

Specifically, I'm having some trouble understanding how to tell Coqui to expect words to appear in specific orders or set weights depending on the situation.

For instance, after asking a yes/no question, I'd like the computer to expect a yes or no answer, but also be able to recognize when an utterance is outside that domain. I'd like to be able to boost the weight of "yes" and "no" so if the computer hears "is peas" it will understand that "is peas" is very unlikely in the situation, but "yes, please" is quite likely.

If it hears "Tell me what time it is" it can recognize that the response is not a simple yes/no response, and treat it as a new conversation.

Later, if the computer were to ask "what is your favorite garden food?" it might understand "is peas" as "it's peas".

I'm using KenLM to generate a language model, but from what I have read, it sounds like I would be better off with a grammar due to the small size of my training data.

A few words to point me in the right direction would be greatly appreciated.

Thank you,
Aaron