PURPOSE: To obtain the speech encoder decoder that enables an adaptive code book to follow up even a part which is not seen in a past sound source signal pattern like the rising of a speech and generates no speech decision error.

CONSTITUTION: This device is equipped with a speech encoder including a means 7 which outputs the indexes of an optimum code book, having the least error based upon a source speech signal among prediction signals, in the lump and means 15 and 16 which find a residue from information on an analytic frame represented by the indexes of the optimum code book and information on a frame abutting in time and encodes and outputs the residue, and a speech decoder including a means which receives the encoded code from the speech encoder and separates information on the analytic frame and information on a frame abutting in time and a means 18 which decodes the information abutting in time into residue information and obtained the adaptive code book from the decoded information.