| 摘要 |
PROBLEM TO BE SOLVED: To provide a speech coding system that has excellent speech quality under a bit rate of 8 kb/s, particularly under 4.8 kb/s in spite of a comparatively small arithmetic quantity and memory capacity. SOLUTION: A spectral parameter calculation section 150 divides a received speech signal into frames with a predetermined time length to obtain a spectrum parameter denoting a spectrum envelope for dividing the frame into sub-frames with a predetermined time length. An adaptive code book section 160 obtains a parameter in an adaptive code book on the basis of a past speech source signal by each sub frame, and a speech source code book retrieval section 170 in speech source code books of which comprise multi-stage cascade connection of a plurality of code books, at least one of the stages outputs a plurality of object code vectors in the order of smaller quantization distortion, and outputs a set of objects minimizing accumulated distortion resulting from accumulating distortions in the entire stage number determined in advance or minimizing quantization distortion of the final stage to quantize the speech signal. |
