The invention relates to an information hiding method for linear prediction voice coding. The method includes the following steps: regarding all positions, at which embedment can be carried out through use of QIM, in a linear prediction voice bit stream as an information hiding space and representing the space with an array A; according to a needed security level, selecting an embedment rate; partitioning the array A so as to obtain a partitioning array; according to a p and the embedment rate, generating a Key and generating a to-be-embedded array B; partitioning the to-be-embedded array B so as to obtain minimum embedment units C; according to the Key, selecting an embedment mode in the minimum embedment units so as to obtain one or more to-be-embedded sequences, wherein each to-be-embedded sequence includes a plurality of embedment positions; and on the basis of the to-be-embedded sequences, realizing hiding of secret information through combination of matrix coding and a QIM (Quantization Index Modulation) method. The invention also provides an information extracting method corresponding to the information hiding method.