Utilizing state-level distance vector representation for improved spoken term detection by text and spoken queries.