Large-vocabulary speaker-independent continuous speech recognition: The SPHINX system