Pause without having to load model again

Another question: I would love to pause recognition and resume at a later time without having to re-load the model. This may take too much time over a mobile data line. I currently stop and restart the audioTracks and that seems to be working, but I get occasional "result" callbacks from the recognizer. Any suggestion?