PURPOSE: To provide the speech encoding method which can obtain a synthesized speech of high quality even at a low bit rate of ≤4kbps and is relatively small in throughput.
CONSTITUTION: A CELP encoder is equipped with an acoustic classifying unit 15, plural statistical code books 32a...32m, and a statistical code book selector 33, and those statistical code books are switched by the statistical code book selector 33 according to the classification result of the acoustic classifying unit 15 and retrieved by a code book retrieving unit 31. Therefore, the statistical code books 32a-32m are generated by using learning data classified previously by the acoustic classifying unit, so a variation of input speeches can be covered and the quality is improved. Further, only the selected code book is used as a code book retrieval at the time of encoding, so the throughput is reducible.
Hidetoshi Sekine
Yasuko Shinada