A voice file retrieval method comprising the steps of: inputting a word; determining if voicing of the word is needed or not; obtaining a storage home address of a voice file corresponding to the word from a voice field of the word if the voice of the word is needed; and retrieving the voice file from the storage home address. By providing a voice field of the word, the storage home address of the voice file can be directly obtained. Hence, the retrieval speed can be increased, and the time to wait for an articulation of the word can be shortened.