To provide a voice/music sound coding device that can improve the coding performance of a voice signal.

A backward filter circuit 10 expresses a reproduced sound source candidate signal combined by a pitch vector from a pitch filter buffer circuit 8, a code vector from a sound source code book circuit 16, and a gain value from a gain code book circuit 17 with a linear prediction coefficient indicating a minute spectrum shape. A synthesizing filter circuit 11 expresses a reproduced sound source candidate signal with a linear prediction coefficient indicating a spectrum outline. An error evaluating circuit 14 calculates the sum of squares of difference signals calculated by a difference circuit 13 for each combination of a pitch vector, a code vector, and a gain value, and outputs indexes corresponding to a pitch vector, a code vector, and a gain value of which the sum of squares is the minimum value from a multiplexer 20.