Large-vocabulary speaker-independent continuous speech recognition: the sphinx system