Combination of a Hidden Tag Model and a Traditional N-gram Model: A Case Study in Czech Speech Recognition