| 摘要 |
PURPOSE: To optimize the discriminative origin vectors of phonemes to output a long sentence with a small memory capacity by using discriminative origin, which are optimized by introducing inclusion relations, namely, subordinate occurrence and exclusiveness, namely, asynchronous occurrence of discriminative origin of phonemes, based of features specific to the Japanese language. CONSTITUTION: A KANA (Japanese syllabary) character string with phonemic symbols inputted from an input terminal 1 is converted into a phonemic symbol string and discriminative origin vectors belonging to it in accordance with standard processing procedures by a phonemic symbol giving circuit 2. A phoneme connection processing circuit 3 corrects the phonemic symbol string and discriminative origin vectors in accordance with the mode of a chain of phonemes. Processings performed in the circuit 3 are silencing processing of vowels, articulation point correction of syllabic nasals, etc. Discriminative origins generated by circuits 2 and 3 are originally vectors of 60 or more characters, but they can be compressed to vectors of about 10 characters in consideration of features specific to the Japanese language. Thus, the constitution of discriminative origin vectors is optimized to process a long sentence with a small memory capacity, and it is outputted with voice. |
