Very glad to hear it was just an issue with stopping. I misunderstood your application to be something like this, where a timed stop is being used to basically interrupt the user mid-utterance and force recognition (an utterance is a continuous period of user speech):
This design (the push-to-talk design mentioned in the linked thread) is trying to sort of fake OpenEars into not being a continuous listener, which would be better solved by using a much more basic Pocketsphinx implementation rather than trying to get OpenEars to be something it’s not with extra code that adds complexity.
What you are talking about sounds a bit different – you are using the continuous listening capabilities, and the user can use the session to speak complete utterances, but you have some kind of arbitrary end to the overall listening session period. I’m going to assume that there is a strong rationale for that design in your app requirements which is why you don’t set the period of listening purely based on user input. I wouldn’t expect your setup to be a problem since your stop is essentially similar to the phone giving an interruption.