摘要 |
PURPOSE:To decrease the operation quantity of matching and to raise the retrieval accuracy by bringing a key of a voice and a key word to matching by a phoneme symbol before comparing them by the respective feature data. CONSTITUTION:A key of an input voice is brought to acoustic analysis 7 and feature data is determined, and a phoneme lattice is recognized 9. Subsequently, from the key, a section like a word is detected 10, and its phoneme lattice is held in a word section phoneme lattice file 11. The whole key is compressed and encoded 4, and held in a file 5. Next, when a retrieval of a data base is started, a hypothesizing part 16 compares a keyword inputted by a phoneme symbol and the phoneme lattice of the word section in the file 11 and sets a candidate of existence of the key word, and a verifying part 19 compares a result of analysis of the key word by a voice and feature data of a key word for a word with respect to its candidate and verifies existence of the key word. Thereafter, a decoding part 23 decodes the key of the voice from the file 5, and a leading-out part 27 outputs data of various forms of a voice, an image, a character-string, etc., being incidental thereto. |