To provide a method and a device for voice coding in which the bit rate is low and the delay is made small while the quality of the reproduced voice can be enhanced.

The device is provided with code tables 111 and 112 which store code vectors for the vector quantization of voice signals. Reproduced voice vectors 104 are generated using the code vectors taken out from the tables. Then, error vectors 106 are generated representing the errors of the reproduced voice vectors against target vectors 102 which are the inputted voice signals. Then, weighted error vectors 108 are generated from the error vectors 106 through a hearing weighted filter 107 having the transfer function which includes the inverse characteristics of the transfer function of a post filter. The code vectors, in which the weighted error vectors become minimum, are searched from the tables 111 and 112 by a code table searching section 109 and the indexes corresponding to the searched code vectors are outputted from a multiplexer 117 as coded parameters.