Initially take a long time to load model.phonemized is poor than trained data.sometime strange speak would happen