I've been doing some expirementations on sphinx-3.5. The batch decoder works great, however, when I try the same model with the live decoder, results get very poor. I understand that the live decoder will surely use a faster search algorithm, but I still think there should be a way to improve it by playing with it's parameters.
My problem is that I can't find any documentation that explains how the live decoder works, does anyone have a clue?