Using words and phonetic strings for efficient information retrieval from imperfectly transcribed spoken documents