Breadth-first search for finding the optimal phonetic transcription from multiple utterances.