Using machine learning method and subword unit representations for spoken document categorization.