Intensive acoustic models constructed by integrating low-occurrence models for spoken term detection.